Models · 479 open models · 12 categories · news checked 2026-10-27What it costs to run an open model
Every open-weight model of the catalogue, from the newest frontier MoE to image, video, speech and embedding models: sized from its weights at FP16, FP8 and INT4, then matched to the smallest real GPU configuration on the board, priced from today's fresh on-demand offers. Monthly figures assume 730 hours.
Models47996 publishers
Categories12chat, code, vision, OCR, image, video, speech, music, 3D, embeddings
Fit on one GPU324at FP8 or INT4, on the board
Throughput estimates412rows, internal study, 2026-10-04
Dated releases140latest Unlimited-OCR-2, Oct 27, 2026
Newest and announced · documented dates · checked 2026-10-27What just shipped
Releases and updates with a documented date, newest first. Announced models stay here, without sizing, until their weights are public.
Oct 27, 2026new
Unlimited-OCR-2Baidu · Vision · 3.34BOCR / document-parsing VLM. README news: '[2026/10/27] We release Unlimited OCR 2'. Small MoE with 64 experts, 6 active. head_dim = 1280/10. Version 1 had about 2.8M downloads.
Oct 24, 2026new
Nanbeige4.3-33BNanbeige · Chat and reasoning · 33.1BDense model (Llama architecture), 512K context, aimed at agentic coding and office tasks. Card calls it 'the flagship teacher model for the Nanbeige4.3 series'; a 3B student is planned but not released. Base model published at Nanbeige/Nanbeige4.3-33B-Base. HF first commit 2026-10-24. #3 on HF trending on 2026-10-27. daily.dev describes Nanbeige as the research team of a job-recruitment app.
Oct 23, 2026new
MiniMax-M3.1-FlashMiniMax · Chat and reasoning · 428B (23B active)Hosted launch 2026-10-19; weights 2026-10-23 (HF first commit and TechNode). Card: 'Approximately 428B total; 23B activated'; HF counts 220.5B because FP4 values are packed. Text, image and video input; context up to 1,000,000 tokens per the card. It keeps M3's MoE architecture, with MiniMax Sparse Attention in all layers, QAT, and a DSpark draft head (an extra 2.30 GB in dspark/). Download is about 3.4x smaller than M3 in BF16 (854 GB).
Oct 23, 2026newDerivative: card says it is 'further post-trained from Gemma-4-E4B-it by NVIDIA', with the same architecture and memory profile as Gemma 4 E4B. Text in, text out; 64K input and 64K output. Card 'Release Date: Hugging Face 10/23/2026'. The 4GB repo was created 2026-10-05 and was trending earlier.
Oct 23, 2026new
Atria DawnShanghai AI Lab · Chat and reasoning · 753.3BAgentic model 'built on the 744B-parameter MoE GLM-5.2 foundation model' (card), with the same memory profile as GLM-5.2. The README table lists a 512K context for the full release; config max_position_embeddings is 1,048,576. The full release replaces Atria Dawn Preview (MIT, 256K, HF 2026-09-11). Text-only. HF first commit 2026-10-23. One October secondary source still listed the full BF16/FP8 downloads as 'forthcoming'; they are now on HF. MLA kv_lora 512. The config head_dim (64) differs from the Preview's 192.
Oct 23, 2026announcedReleased on 2026-10-23 in Moonshot's apps and API (a Highspeed variant exists; API prices $2 / $10 per million tokens, $4 / $20 for Highspeed). A post-trained version of Kimi K3: 2.8T parameters, 104B active, 1M-token context. Weights are not downloadable yet: Moonshot's blog says the full weights will be released by 2026-11-04; no licence announced, and no K3.1 repository on Hugging Face on 2026-10-27. Published scores (65.7% on Terminal-Bench 4.0) are vendor-declared. No model simply named 'Kimi K3.1' exists.
Oct 23, 2026new
Hi3D v1.0Hi3D · 3DRefines a coarse mesh into high-resolution geometry conditioned on an image (geometry only, no texture); trained on about 2M assets. Files: transformer_512 3.68 GB + transformer_1024 3.68 GB + sparse VAE 0.38 GB. Uses Hunyuan3D-2mini for coarse shapes in examples. Parameter count not stated.
Oct 21, 2026new
Raon-1 miniKRAFTON · Chat and reasoning · 28.8B (4.3B active)Preview release, trained from scratch and strong in Korean. Card: 'Total Parameters 28.8B, Activated Parameters 4.3B', 25 layers = 18 Raon Delta Attention (linear) + 7 MLA. 'Only the 7 MLA layers keep a KV cache, about 8 KiB per token in BF16'. 128 experts, 6 active, plus 1 shared. NVFP4 build fits a single RTX 5090 (card). HF first commit 2026-10-21; only one secondary listing was found.
The whole catalogueSized and priced, by category
Search by name, publisher, repository or licence; pick a category. Each row links to the model's page with every configuration and the study's throughput estimates.
479 models · sized at the most useful precision we could compute (FP8 when the weights allow it, INT4 for the largest) · smallest configuration among 1, 2, 4 and 8 GPUs of one type, priced from the cheapest fresh on-demand offer
Chat and reasoning 211
| Model | Params | Checkpoint | Context | Licence | Needs | Smallest config | $ / hour | $ / month | Release | |
|---|---|---|---|---|---|---|---|---|---|---|
 Qwen3 0.6BQwen · Qwen3-0.6B | 752M | 2 GBBF16 | 40k | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Qwen3 0.6B GPU requirements |
| Qwen3 14BQwen · Qwen3-14B | 14.8B | 30 GBBF16 | 40k | apache-2.0 | 18 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | Qwen3 14B GPU requirements |
| Qwen3 4BQwen · Qwen3-4B | 4B | 8 GBBF16 | 40k | apache-2.0 | 6 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Qwen3 4B GPU requirements |
 Gemma 4 31BGoogle · gemma-4-31B-it | 31.3B | 63 GBBF16 | 256k | apache-2.0 | 42 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | Apr 2, 2026 | Gemma 4 31B GPU requirements |
| Qwen3 8BQwen · Qwen3-8B | 8.2B | 16 GBBF16 | 40k | apache-2.0 | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Qwen3 8B GPU requirements |
| Qwen3.8 27BQwen · Qwen3.8-27B | 27.8B | 56 GBBF16 | 256k | apache-2.0 | 34 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | Aug 2026 | Qwen3.8 27B GPU requirements |
| Qwen3.5 4BQwen · Qwen3.5-4B | 4.66B | 9 GBBF16 | 256k | apache-2.0 | 6 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Qwen3.5 4B GPU requirements |
| gemma 4 26B A4B itGoogle · gemma-4-26B-A4B-it | 25.8B · 4B active | 52 GBBF16 | 256k | apache-2.0 | 31 GBFP8 (est.) | 1× Tesla V100 32GBat Vast.ai | $0.07 | $50 | Apr 2, 2026 | gemma 4 26B A4B it GPU requirements |
 GLM-5.3 FlashZ.ai · GLM-5.3-Flash | 321B · 18B active | 328 GBFP8 | 1M | mit | 394 GBFP8 | 2× MI325Xat Cyfuture AI | $6.21 | $4,536 | Aug 25, 2026 | GLM-5.3 Flash GPU requirements |
| Qwen3.5 9BQwen · Qwen3.5-9B | 9.65B | 19 GBBF16 | 256k | apache-2.0 | 12 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Qwen3.5 9B GPU requirements |
 DeepSeek V3.2DeepSeek · DeepSeek-V3.2 | 685B · 37B active | 689 GBFP8 | 160k | mit | 828 GBFP8 | 4× MI325Xat Cyfuture AI | $12.29 | $8,972 | — | DeepSeek V3.2 GPU requirements |
 gpt-oss-20bOpenAI | 21B · 3.6B active | 14 GBINT4-packed + BF16 | 128k | apache-2.0 | 17 GBINT4-packed + BF16 | 1× Tesla P40at Clore.ai | $0.04 | $30 | Aug 5, 2025 | gpt-oss-20b GPU requirements |
 Llama 3.1 8B InstructMeta · Llama-3.1-8B-Instruct | 8B | 16 GBBF16 | 128k | llama3.1 | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Llama 3.1 8B Instruct GPU requirements |
| Qwen3.6 35B-A3BQwen · Qwen3.6-35B-A3B | 35B · 3B active | 72 GBBF16 | 256k | apache-2.0 | 48 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Qwen3.6 35B-A3B GPU requirements |
| Qwen3 1.7BQwen · Qwen3-1.7B | 2.03B | 4 GBBF16 | 40k | apache-2.0 | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Qwen3 1.7B GPU requirements |
| gemma 4 E4B itGoogle · gemma-4-E4B-it | 8B · 4.5B active | 16 GBBF16 | 128k | apache-2.0 | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Apr 2, 2026 | gemma 4 E4B it GPU requirements |
| Qwen3.5 0.8BQwen · Qwen3.5-0.8B | 873M | 2 GBBF16 | 256k | apache-2.0 | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Qwen3.5 0.8B GPU requirements |
| Qwen3.8 Flash NextQwen · Qwen3.8-Flash-Next | 180B · 6B active | 360 GBBF16 | 256k | other | 216 GBFP8 (est.) | 8× Tesla V100 32GBat Vast.ai | $0.54 | $391 | — | Qwen3.8 Flash Next GPU requirements |
| DeepSeek V4 Flash 0731DeepSeek · DeepSeek-V4-Flash-0731 | 304B · 13B active | 167 GBINT4-packed | 1M | mit | 201 GBINT4-packed | 8× Tesla V100 32GBat Vast.ai | $0.54 | $391 | Jul 31, 2026 | DeepSeek V4 Flash 0731 GPU requirements |
| Qwen3.5 2BQwen · Qwen3.5-2B | 2.27B | 5 GBBF16 | 256k | apache-2.0 | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Qwen3.5 2B GPU requirements |
| gemma 4 E2B itGoogle · gemma-4-E2B-it | 5.12B · 2.3B active | 10 GBBF16 | 128k | apache-2.0 | 7 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Apr 2, 2026 | gemma 4 E2B it GPU requirements |
| Llama 3.2 1B InstructMeta · Llama-3.2-1B-Instruct | 1.24B | 3 GBBF16 | 128k | llama3.2 | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Llama 3.2 1B Instruct GPU requirements |
| Qwen3 32BQwen · Qwen3-32B | 32.8B | 66 GBBF16 | 40k | apache-2.0 | 44 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Qwen3 32B GPU requirements |
| gpt-oss-120bOpenAI | 117B · 5.1B active | 65 GBINT4-packed | 128k | apache-2.0 | 79 GBINT4-packed | 4× RTX 3090at Clore.ai | $0.25 | $183 | Aug 5, 2025 | gpt-oss-120b GPU requirements |
| Qwen3 4B Instruct 2507Qwen · Qwen3-4B-Instruct-2507 | 4.02B | 8 GBBF16 | 256k | apache-2.0 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Qwen3 4B Instruct 2507 GPU requirements |
| Qwen3.6 27BQwen · Qwen3.6-27B | 27.8B | 56 GBBF16 | 256k | apache-2.0 | 37 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Qwen3.6 27B GPU requirements |
| GLM-5.3Z.ai | 753B · 40B active | 755 GBFP8 | 1M | GLM-5.3 License | 907 GBFP8 | 4× MI325Xat Cyfuture AI | $12.29 | $8,972 | Aug 27, 2026 | GLM-5.3 GPU requirements |
| InklingThinking Machines | 952B · 41B active | 1,905 GBBF16 | — | apache-2.0 | 1,143 GBFP8 (est.) | 8× MI300Xat TensorWave | $13.68 | $9,986 | Jul 15, 2026 | Inkling GPU requirements |
| gemma 4 12B itGoogle · gemma-4-12B-it | 12B | 24 GBBF16 | 256k | apache-2.0 | 15 GBFP8 (est.) | 1× T4at Vast.ai | $0.04 | $29 | — | gemma 4 12B it GPU requirements |
 Mistral 7B Instruct v0.3Mistral AI · Mistral-7B-Instruct-v0.3 | 7.2B | 14 GBBF16 | 32k | apache-2.0 | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Mistral 7B Instruct v0.3 GPU requirements |
| DeepSeek R1DeepSeek · DeepSeek-R1 | 671B · 37B active | 689 GBFP8 | 160k | mit | 827 GBFP8 | 4× MI325Xat Cyfuture AI | $12.29 | $8,972 | — | DeepSeek R1 GPU requirements |
| Inkling SmallThinking Machines · Inkling-Small | 266B · 12B active | 532 GBBF16 | — | apache-2.0 | 320 GBFP8 (est.) | 4× A100 80GBat Vast.ai | $2.40 | $1,754 | Jul 30, 2026 | Inkling Small GPU requirements |
| MiniCPM5 2BOpenBMB · MiniCPM5-2B | 2.52B | 5 GBBF16 | 128k | apache-2.0 | 4 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | MiniCPM5 2B GPU requirements |
| GLM-4.5 AirZ.ai · GLM-4.5-Air | 110B · 12B active | 221 GBBF16 | 128k | mit | 132 GBFP8 (est.) | 8× Tesla V100 32GBat Vast.ai | $0.54 | $391 | — | GLM-4.5 Air GPU requirements |
| Hy3Tencent | 299B · 21B active | 598 GBBF16 | 256k | apache-2.0 | 359 GBFP8 (est.) | 8× RTX A6000at Lium | $3.36 | $2,453 | — | Hy3 GPU requirements |
| Llama 3.2 3B InstructMeta · Llama-3.2-3B-Instruct | 3.21B | 6 GBBF16 | 128k | llama3.2 | 4 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Llama 3.2 3B Instruct GPU requirements |
| Qwen3 30B A3B Instruct 2507Qwen · Qwen3-30B-A3B-Instruct-2507 | 30.5B · 3B active | 61 GBBF16 | 256k | apache-2.0 | 37 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Qwen3 30B A3B Instruct 2507 GPU requirements |
| DeepSeek R1 Distill Qwen 1.5BDeepSeek · DeepSeek-R1-Distill-Qwen-1.5B | 1.78B | 4 GBBF16 | 128k | mit | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | DeepSeek R1 Distill Qwen 1.5B GPU requirements |
| DeepSeek V4 FlashDeepSeek · DeepSeek-V4-Flash | 284B | 160 GBINT4-packed | 1M | mit | 192 GBINT4-packed | 8× Tesla V100 32GBat Vast.ai | $0.54 | $391 | Apr 24, 2026 | DeepSeek V4 Flash GPU requirements |
| GLM-4.7 FlashZ.ai · GLM-4.7-Flash | 31.2B | 62 GBBF16 | 198k | mit | 42 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | GLM-4.7 Flash GPU requirements |
| Qwen3.5 122B A10BQwen · Qwen3.5-122B-A10B | 125B · 10B active | 250 GBBF16 | 256k | apache-2.0 | 150 GBFP8 (est.) | 8× Tesla V100 32GBat Vast.ai | $0.54 | $391 | — | Qwen3.5 122B A10B GPU requirements |
| Qwen3.5 397B A17BQwen · Qwen3.5-397B-A17B | 403B · 17B active | 807 GBBF16 | 256k | apache-2.0 | 484 GBFP8 (est.) | 2× MI325Xat Cyfuture AI | $6.21 | $4,536 | Feb 16, 2026 | Qwen3.5 397B A17B GPU requirements |
| Kimi K3Moonshot AI · Kimi-K3 | 2.8T · 104B active | 1,560 GBINT4-packed | 1M | Kimi K3 License | 1,874 GBINT4-packed | 8× MI325Xat TensorWave | $18.00 | $13,140 | Jul 27, 2026 | Kimi K3 GPU requirements |
| SmolLM3 3BHugging Face · SmolLM3-3B | 3.08B | 6 GBBF16 | 64k | apache-2.0 | 4 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | SmolLM3 3B GPU requirements |
| diffusiongemma 26B A4B itGoogle · diffusiongemma-26B-A4B-it | 25.8B · 4B active | 52 GBBF16 | 256k | apache-2.0 | 31 GBFP8 (est.) | 1× Tesla V100 32GBat Vast.ai | $0.07 | $50 | — | diffusiongemma 26B A4B it GPU requirements |
| Qwen3 Omni 30B A3B InstructQwen · Qwen3-Omni-30B-A3B-Instruct | 35.3B · 3B active | 71 GBBF16 | — | other | 43 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Qwen3 Omni 30B A3B Instruct GPU requirements |
| Qwen3 4B Thinking 2507Qwen · Qwen3-4B-Thinking-2507 | 4.02B | 8 GBBF16 | 256k | apache-2.0 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Qwen3 4B Thinking 2507 GPU requirements |
 granite-4.1-3bIBM | 3.4B | 7 GBBF16 | 128k | apache-2.0 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | granite-4.1-3b GPU requirements |
| MiniMax M2.7MiniMax · MiniMax-M2.7 | 229B · 10B active | 230 GBFP8 | 200k | other | 277 GBFP8 | 4× A100 80GBat Vast.ai | $2.40 | $1,754 | — | MiniMax M2.7 GPU requirements |
| Llama 3.3 70B InstructMeta · Llama-3.3-70B-Instruct | 70.6B | 141 GBBF16 | 128k | llama3.3 | 85 GBFP8 (est.) | 4× RTX 3090at Clore.ai | $0.25 | $183 | — | Llama 3.3 70B Instruct GPU requirements |
| Qwen3 235B-A22BQwen · Qwen3-235B-A22B | 235B · 22B active | 470 GBBF16 | 40k | apache-2.0 | 311 GBFP8 (est.) | 4× A100 80GBat Vast.ai | $2.40 | $1,754 | — | Qwen3 235B-A22B GPU requirements |
 phi-4Microsoft | 14.7B | 29 GBBF16 | 16k | mit | 18 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | phi-4 GPU requirements |
| Kimi K2.5Moonshot AI · Kimi-K2.5 | 1T · 32B active | 595 GBINT4-packed | 256k | modified-mit | 715 GBINT4-packed | 8× RTX PRO 6000at UpCloud | $7.78 | $5,676 | — | Kimi K2.5 GPU requirements |
| DeepSeek V4 ProDeepSeek · DeepSeek-V4-Pro | 1.6T · 49B active | 865 GBINT4-packed | 1M | mit | 1,038 GBINT4-packed | 8× MI300Xat TensorWave | $13.68 | $9,986 | Aug 13, 2026 | DeepSeek V4 Pro GPU requirements |
| DeepSeek R1 0528 Qwen3 8BDeepSeek · DeepSeek-R1-0528-Qwen3-8B | 8.19B | 16 GBBF16 | 128k | mit | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | DeepSeek R1 0528 Qwen3 8B GPU requirements |
| DeepSeek R1 Distill Qwen 32BDeepSeek · DeepSeek-R1-Distill-Qwen-32B | 32.8B | 66 GBBF16 | 128k | mit | 40 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | DeepSeek R1 Distill Qwen 32B GPU requirements |
| Phi 4 mini instructMicrosoft · Phi-4-mini-instruct | 3.84B | 8 GBBF16 | 128k | mit | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Phi 4 mini instruct GPU requirements |
| Gemma 3 12BGoogle · gemma-3-12b-it | 12.2B | 24 GBBF16 | 128k | gemma | 17 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | Gemma 3 12B GPU requirements |
| gemma-3-27b-itGoogle | 27.4B | 55 GBBF16 | 128k | gemma | 33 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | gemma-3-27b-it GPU requirements |
| Qwen3 Omni 30B A3B ThinkingQwen · Qwen3-Omni-30B-A3B-Thinking | 31.7B · 3B active | 63 GBBF16 | — | other | 39 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Qwen3 Omni 30B A3B Thinking GPU requirements |
| DeepSeek R1 Distill Qwen 7BDeepSeek · DeepSeek-R1-Distill-Qwen-7B | 7.62B | 15 GBBF16 | 128k | mit | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | DeepSeek R1 Distill Qwen 7B GPU requirements |
| MiniCPM5 1BOpenBMB · MiniCPM5-1B | 1.08B | 2 GBBF16 | 128k | apache-2.0 | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | MiniCPM5 1B GPU requirements |
| Kimi K2.6Moonshot AI · Kimi-K2.6 | 1T · 32B active | 595 GBINT4-packed | 256k | other | 715 GBINT4-packed | 8× RTX PRO 6000at UpCloud | $7.78 | $5,676 | — | Kimi K2.6 GPU requirements |
| DeepSeek V3.1DeepSeek · DeepSeek-V3.1 | 685B · 37B active | 689 GBFP8 | 160k | mit | 827 GBFP8 | 4× MI325Xat Cyfuture AI | $12.29 | $8,972 | — | DeepSeek V3.1 GPU requirements |
| GLM-5.2Z.ai | 753B · 40B active | 1,507 GBBF16 | 1M | mit | 995 GBFP8 (est.) | 4× MI325Xat Cyfuture AI | $12.29 | $8,972 | Jun 16, 2026 | GLM-5.2 GPU requirements |
| DeepSeek R1 Distill Qwen 14BDeepSeek · DeepSeek-R1-Distill-Qwen-14B | 14.8B | 30 GBBF16 | 128k | mit | 18 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | DeepSeek R1 Distill Qwen 14B GPU requirements |
| Step 3.5 FlashStepFun · Step-3.5-Flash | 199B · 11B active | 399 GBBF16 | 256k | apache-2.0 | 239 GBFP8 (est.) | 8× Tesla V100 32GBat Vast.ai | $0.54 | $391 | — | Step 3.5 Flash GPU requirements |
| MiniCPM o 4.5OpenBMB · MiniCPM-o-4_5 | 9.37B | 19 GBBF16 | 40k | apache-2.0 | 12 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | MiniCPM o 4.5 GPU requirements |
| Apertus v1.5 8BSwiss AI · Apertus-v1.5-8B | 8.9B | 18 GBBF16 | — | apache-2.0 | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Apertus v1.5 8B GPU requirements |
| Olmo 3 7B InstructAi2 · Olmo-3-7B-Instruct | 7.3B | 15 GBBF16 | 64k | apache-2.0 | 9 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Olmo 3 7B Instruct GPU requirements |
| Ministral 3 14B Reasoning 2512Mistral AI · Ministral-3-14B-Reasoning-2512 | 13.9B | 28 GBBF16 | 256k | apache-2.0 | 17 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | Ministral 3 14B Reasoning 2512 GPU requirements |
| Muse Glimmer 30BMeta · Muse-Glimmer-30B | 29.8B | 60 GBBF16 | 128k | apache-2.0 | 36 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | Aug 10, 2026 | Muse Glimmer 30B GPU requirements |
| MiniMax M3MiniMax · MiniMax-M3 | 427B · 23B active | 854 GBBF16 | 1M | other | 513 GBFP8 (est.) | 8× RTX PRO 6000at UpCloud | $7.78 | $5,676 | — | MiniMax M3 GPU requirements |
| NVIDIA Nemotron Nano 9B v2NVIDIA · NVIDIA-Nemotron-Nano-9B-v2 | 8.89B | 18 GBBF16 | 128k | other | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | NVIDIA Nemotron Nano 9B v2 GPU requirements |
| DeepSeek R1 0528DeepSeek · DeepSeek-R1-0528 | 685B · 37B active | 689 GBFP8 | 160k | mit | 827 GBFP8 | 4× MI325Xat Cyfuture AI | $12.29 | $8,972 | — | DeepSeek R1 0528 GPU requirements |
| DeepSeek V4 Pro 0813DeepSeek · DeepSeek-V4-Pro-0813 | 1.6T · 49B active | 893 GBINT4-packed | 1M | mit | 1,072 GBINT4-packed | 8× MI300Xat TensorWave | $13.68 | $9,986 | Aug 13, 2026 | DeepSeek V4 Pro 0813 GPU requirements |
| Mistral Small 3.2 24BMistral AI · Mistral-Small-3.2-24B-Instruct-2506 | 24B | 48 GBBF16 | 128k | apache-2.0 | 32 GBFP8 (est.) | 1× Tesla V100 32GBat Vast.ai | $0.07 | $50 | — | Mistral Small 3.2 24B GPU requirements |
| Ministral 3 14B Instruct 2512Mistral AI · Ministral-3-14B-Instruct-2512 | 13.9B | 16 GBFP8 + BF16 | 256k | apache-2.0 | 19 GBFP8 + BF16 | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | Ministral 3 14B Instruct 2512 GPU requirements |
| Qwen3 Next 80B A3B InstructQwen · Qwen3-Next-80B-A3B-Instruct | 81.3B · 3B active | 163 GBBF16 | 256k | apache-2.0 | 98 GBFP8 (est.) | 4× Tesla V100 32GBat Vast.ai | $0.27 | $196 | — | Qwen3 Next 80B A3B Instruct GPU requirements |
| Qwen3 235B A22B Instruct 2507Qwen · Qwen3-235B-A22B-Instruct-2507 | 235B · 22B active | 470 GBBF16 | 256k | apache-2.0 | 282 GBFP8 (est.) | 4× A100 80GBat Vast.ai | $2.40 | $1,754 | — | Qwen3 235B A22B Instruct 2507 GPU requirements |
| LFM2.5-1.2B InstructLiquid AI · LFM2.5-1.2B-Instruct | 1.17B | 2 GBBF16 | 125k | other | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | LFM2.5-1.2B Instruct GPU requirements |
| Kimi Linear 48B A3B InstructMoonshot AI · Kimi-Linear-48B-A3B-Instruct | 49.1B · 3B active | 98 GBBF16 | — | mit | 59 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Kimi Linear 48B A3B Instruct GPU requirements |
| Nanbeige4.2 3BNanbeige · Nanbeige4.2-3B | 4.17B | 8 GBBF16 | 256k | apache-2.0 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Nanbeige4.2 3B GPU requirements |
| DeepSeek V4.1 FlashDeepSeek · DeepSeek-V4.1-Flash | 763.2B · 16B active | 510 GBINT4-packed + FP8/BF16 | 1M | mit | 613 GBINT4-packed + FP8/BF16 | 8× RTX PRO 6000at UpCloud | $7.78 | $5,676 | Oct 17, 2026 | DeepSeek V4.1 Flash GPU requirements |
| Llama 4 Scout 17B 16E InstructMeta · Llama-4-Scout-17B-16E-Instruct | 109B · 17B active | 217 GBBF16 | 10M | Llama 4 Community License | 131 GBFP8 (est.) | 8× Tesla V100 32GBat Vast.ai | $0.54 | $391 | Apr 5, 2025 | Llama 4 Scout 17B 16E Instruct GPU requirements |
| DeepSeek R1 Distill Llama 8BDeepSeek · DeepSeek-R1-Distill-Llama-8B | 8.03B | 16 GBBF16 | 128k | mit | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | DeepSeek R1 Distill Llama 8B GPU requirements |
| LFM2.5-2.6BLiquid AI | 2.7B | 5 GBBF16 | 128k | other | 4 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | LFM2.5-2.6B GPU requirements |
| Granite 4.1 8BIBM · granite-4.1-8b | 8.8B | 18 GBBF16 | 128k | apache-2.0 | 12 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Granite 4.1 8B GPU requirements |
| Ministral 3 8B Instruct 2512Mistral AI · Ministral-3-8B-Instruct-2512 | 8.92B | 10 GBFP8 + BF16 | 256k | apache-2.0 | 13 GBFP8 + BF16 | 1× T4at Vast.ai | $0.04 | $29 | — | Ministral 3 8B Instruct 2512 GPU requirements |
| LFM2.5-350MLiquid AI | 354M | 1 GBBF16 | 125k | other | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | LFM2.5-350M GPU requirements |
| Ministral 3 3B Instruct 2512Mistral AI · Ministral-3-3B-Instruct-2512 | 3.85B | 5 GBFP8 + BF16 | 256k | apache-2.0 | 6 GBFP8 + BF16 | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Ministral 3 3B Instruct 2512 GPU requirements |
| MiMo V2.6 Flash RLXiaomi · MiMo-V2.6-Flash-RL | 311B · 15B active | 178 GBINT4-packed | 1M | mit | 214 GBINT4-packed | 8× Tesla V100 32GBat Vast.ai | $0.54 | $391 | Sep 22, 2026 | MiMo V2.6 Flash RL GPU requirements |
| Llama 3.3 Nemotron Super 49B v1.5NVIDIA · Llama-3_3-Nemotron-Super-49B-v1_5 | 49.9B | 100 GBBF16 | 128k | other | 60 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Llama 3.3 Nemotron Super 49B v1.5 GPU requirements |
| GLM-4.7Z.ai | 358B · 32B active | 717 GBBF16 | 198k | mit | 430 GBFP8 (est.) | 2× MI325Xat Cyfuture AI | $6.21 | $4,536 | — | GLM-4.7 GPU requirements |
| Mistral Small 4 119B 2603Mistral AI · Mistral-Small-4-119B-2603 | 119B · 6.5B active | 121 GBFP8 | 1M | apache-2.0 | 146 GBFP8 | 8× Tesla V100 32GBat Vast.ai | $0.54 | $391 | Mar 16, 2026 | Mistral Small 4 119B 2603 GPU requirements |
| Hunyuan A13B InstructTencent · Hunyuan-A13B-Instruct | 80.4B · 13B active | 161 GBBF16 | 32k | other | 97 GBFP8 (est.) | 4× Tesla V100 32GBat Vast.ai | $0.27 | $196 | — | Hunyuan A13B Instruct GPU requirements |
| ERNIE-4.5 21B A3B ThinkingBaidu · ERNIE-4.5-21B-A3B-Thinking | 21.8B · 3B active | 44 GBBF16 | 128k | apache-2.0 | 27 GBFP8 (est.) | 1× Tesla V100 32GBat Vast.ai | $0.07 | $50 | — | ERNIE-4.5 21B A3B Thinking GPU requirements |
| Qwen3 30B A3B Thinking 2507Qwen · Qwen3-30B-A3B-Thinking-2507 | 30.5B · 3B active | 61 GBBF16 | 256k | apache-2.0 | 37 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Qwen3 30B A3B Thinking 2507 GPU requirements |
| Kimi K2 ThinkingMoonshot AI · Kimi-K2-Thinking | 1T · 32B active | 594 GBINT4-packed | 256k | other | 714 GBINT4-packed | 8× RTX PRO 6000at UpCloud | $7.78 | $5,676 | — | Kimi K2 Thinking GPU requirements |
| internlm3-8b-instructShanghai AI Lab | 8.8B | 18 GBBF16 | 32k | apache-2.0 | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | internlm3-8b-instruct GPU requirements |
| LFM2.5-230MLiquid AI | 230M | 0 GBBF16 | 125k | other | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | LFM2.5-230M GPU requirements |
| Seed OSS 36B InstructByteDance Seed · Seed-OSS-36B-Instruct | 36.2B | 72 GBBF16 | 512k | apache-2.0 | 44 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Seed OSS 36B Instruct GPU requirements |
| Qwen3 Next 80B A3B ThinkingQwen · Qwen3-Next-80B-A3B-Thinking | 81.3B · 3B active | 163 GBBF16 | 256k | apache-2.0 | 98 GBFP8 (est.) | 4× Tesla V100 32GBat Vast.ai | $0.27 | $196 | — | Qwen3 Next 80B A3B Thinking GPU requirements |
| Phi 4 mini reasoningMicrosoft · Phi-4-mini-reasoning | 3.84B | 8 GBBF16 | 128k | mit | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Phi 4 mini reasoning GPU requirements |
| granite-4.2-30bIBM | 29.3B | 59 GBBF16 | 128k | apache-2.0 | 36 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | Aug 25, 2026 | granite-4.2-30b GPU requirements |
| EXAONE-4.5 33BLG AI Research · EXAONE-4.5-33B | 34.4B | 69 GBBF16 | 256k | other | 42 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | EXAONE-4.5 33B GPU requirements |
| DeepSeek R1 Distill Llama 70BDeepSeek · DeepSeek-R1-Distill-Llama-70B | 70.6B | 141 GBBF16 | 128k | mit | 85 GBFP8 (est.) | 4× RTX 3090at Clore.ai | $0.25 | $183 | — | DeepSeek R1 Distill Llama 70B GPU requirements |
| MiMo V2.6 Pro RLXiaomi · MiMo-V2.6-Pro-RL | 1T · 42B active | 573 GBINT4-packed | 1M | mit | 689 GBINT4-packed | 8× RTX PRO 6000at UpCloud | $7.78 | $5,676 | Sep 22, 2026 | MiMo V2.6 Pro RL GPU requirements |
| Phi 4 reasoningMicrosoft · Phi-4-reasoning | 14.7B | 29 GBBF16 | 32k | mit | 18 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | Phi 4 reasoning GPU requirements |
| Mistral Medium 3.5 128BMistral AI · Mistral-Medium-3.5-128B | 128B | 134 GBFP8 | 256k | Modified MIT | 161 GBFP8 | 8× Tesla V100 32GBat Vast.ai | $0.54 | $391 | Apr 29, 2026 | Mistral Medium 3.5 128B GPU requirements |
| LFM2.5-8B A1BLiquid AI · LFM2.5-8B-A1B | 8.47B · 1B active | 17 GBBF16 | 125k | other | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | LFM2.5-8B A1B GPU requirements |
| granite-4.0-h-tinyIBM | 6.94B · 1.5B active | 14 GBBF16 | 128k | apache-2.0 | 9 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | granite-4.0-h-tiny GPU requirements |
| Magistral Small 2509Mistral AI · Magistral-Small-2509 | 24B | 48 GBBF16 | 128k | apache-2.0 | 29 GBFP8 (est.) | 1× Tesla V100 32GBat Vast.ai | $0.07 | $50 | — | Magistral Small 2509 GPU requirements |
 Solar Open 100BUpstage · Solar-Open-100B | 103B · 12B active | 205 GBBF16 | 128k | other | 124 GBFP8 (est.) | 4× Tesla V100 32GBat Vast.ai | $0.27 | $196 | — | Solar Open 100B GPU requirements |
| granite-4.0-h-smallIBM | 32.2B · 8.8B active | 64 GBBF16 | 128k | apache-2.0 | 39 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | granite-4.0-h-small GPU requirements |
| granite-4.0-microIBM | 3.4B | 7 GBBF16 | 128k | apache-2.0 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | granite-4.0-micro GPU requirements |
| Qwen3.8 2.4T A95BQwen · Qwen3.8-2.4T-A95B | 2.4T · 95B active | 4,892 GBBF16 | 256k | Qwen3.8 licence (revenue threshold $50M) | 2,880 GBFP8 (est.) | more than 8 GPUs | — | — | Aug 2026 | Qwen3.8 2.4T A95B GPU requirements |
| GLM-4.6Z.ai | 357B · 32B active | 714 GBBF16 | 198k | mit | 429 GBFP8 (est.) | 2× MI325Xat Cyfuture AI | $6.21 | $4,536 | — | GLM-4.6 GPU requirements |
| LFM2.5-24B-A2BLiquid AI | 24B · 2.45B active | 48 GBBF16 | 128k | LFM Open License v1.0 | 29 GBFP8 (est.) | 1× Tesla V100 32GBat Vast.ai | $0.07 | $50 | Oct 13, 2026 | LFM2.5-24B-A2B GPU requirements |
| Kolibri-1Aleph Alpha | 78.1B · 3.46B active | 79 GBFP8 | 256k | Apache-2.0 | 95 GBFP8 | 4× RTX 3090at Clore.ai | $0.25 | $183 | Oct 3, 2026 | Kolibri-1 GPU requirements |
| medgemma-27b-text-itGoogle | 27B | 54 GBBF16 | 128k | other | 33 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | medgemma-27b-text-it GPU requirements |
| Olmo 3.1 32B ThinkAi2 · Olmo-3.1-32B-Think | 32.2B | 65 GBBF16 | 64k | apache-2.0 | 39 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Olmo 3.1 32B Think GPU requirements |
| EXAONE-4.0 32BLG AI Research · EXAONE-4.0-32B | 32B | 64 GBBF16 | 128k | other | 39 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | EXAONE-4.0 32B GPU requirements |
| Ministral 3 3B Reasoning 2512Mistral AI · Ministral-3-3B-Reasoning-2512 | 4.25B | 8 GBBF16 | 256k | apache-2.0 | 6 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Ministral 3 3B Reasoning 2512 GPU requirements |
| Hy4 previewTencent · Hy4-preview | 780B · 49B active | 1,560 GBBF16 | 1M | apache-2.0 | 936 GBFP8 (est.) | 4× MI325Xat Cyfuture AI | $12.29 | $8,972 | Aug 28, 2026 | Hy4 preview GPU requirements |
| LongCat Flash ChatMeituan · LongCat-Flash-Chat | 562B · 27B active | 1,124 GBBF16 | 128k | mit | 675 GBFP8 (est.) | 8× RTX PRO 6000at UpCloud | $7.78 | $5,676 | — | LongCat Flash Chat GPU requirements |
| Step 3.7 FlashStepFun · Step-3.7-Flash | 201B · 16.4B active | 403 GBBF16 | 256k | apache-2.0 | 242 GBFP8 (est.) | 8× Tesla V100 32GBat Vast.ai | $0.54 | $391 | May 29, 2026 | Step 3.7 Flash GPU requirements |
| Nemotron Labs Diffusion 14BNVIDIA · Nemotron-Labs-Diffusion-14B | 13.5B | 27 GBBF16 | 256k | other | 17 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | Nemotron Labs Diffusion 14B GPU requirements |
| Tongyi DeepResearch 30B A3BTongyi Lab · Tongyi-DeepResearch-30B-A3B | 30.5B · 3B active | 61 GBBF16 | 128k | apache-2.0 | 37 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Tongyi DeepResearch 30B A3B GPU requirements |
| MiniMax-M3.1-FlashMiniMax | 428B · 23B active | 250 GBINT4-packed | 1M | MiniMax Community License | 301 GBINT4-packed | 4× A100 80GBat Vast.ai | $2.40 | $1,754 | Oct 23, 2026 | MiniMax-M3.1-Flash GPU requirements |
| Qwen3 235B A22B Thinking 2507Qwen · Qwen3-235B-A22B-Thinking-2507 | 235B · 22B active | 470 GBBF16 | 256k | apache-2.0 | 282 GBFP8 (est.) | 4× A100 80GBat Vast.ai | $2.40 | $1,754 | — | Qwen3 235B A22B Thinking 2507 GPU requirements |
| gemma 3n E4B itGoogle · gemma-3n-E4B-it | 7.85B | 16 GBBF16 | 32k | gemma | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | gemma 3n E4B it GPU requirements |
| Ling 3.0 tinyinclusionAI · Ling-3.0-tiny | 7.89B · 1.3B active | 16 GBBF16 | 128k | mit | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Ling 3.0 tiny GPU requirements |
| tiny-aya-globalCohere | 3.35B | 7 GBBF16 | — | cc-by-nc-4.0 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | tiny-aya-global GPU requirements |
| DeepSeek V3.1 TerminusDeepSeek · DeepSeek-V3.1-Terminus | 685B · 37B active | 689 GBFP8 | 160k | mit | 827 GBFP8 | 4× MI325Xat Cyfuture AI | $12.29 | $8,972 | — | DeepSeek V3.1 Terminus GPU requirements |
| Trinity MiniArcee · Trinity-Mini | 26.1B · 3B active | 52 GBBF16 | 128k | other | 32 GBFP8 (est.) | 1× Tesla V100 32GBat Vast.ai | $0.07 | $50 | — | Trinity Mini GPU requirements |
| K-EXAONE 236B A23BLG AI Research · K-EXAONE-236B-A23B | 237B · 23B active | 474 GBBF16 | 256k | other | 285 GBFP8 (est.) | 4× A100 80GBat Vast.ai | $2.40 | $1,754 | — | K-EXAONE 236B A23B GPU requirements |
| Llama 4 Maverick 17B 128E InstructMeta · Llama-4-Maverick-17B-128E-Instruct | 402B · 17B active | 803 GBBF16 | 1M | Llama 4 Community License | 483 GBFP8 (est.) | 2× MI325Xat Cyfuture AI | $6.21 | $4,536 | Apr 5, 2025 | Llama 4 Maverick 17B 128E Instruct GPU requirements |
| HyperCLOVAX SEED Think 32BNAVER · HyperCLOVAX-SEED-Think-32B | 33.3B | 67 GBBF16 | 128k | other | 40 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | HyperCLOVAX SEED Think 32B GPU requirements |
| Intern S1Shanghai AI Lab · Intern-S1 | 241B · 27.8B active | 481 GBBF16 | 64k | apache-2.0 | 290 GBFP8 (est.) | 4× A100 80GBat Vast.ai | $2.40 | $1,754 | — | Intern S1 GPU requirements |
| Ling 3.0 flashinclusionAI · Ling-3.0-flash | 127B · 5.1B active | 255 GBBF16 | 256k | mit | 153 GBFP8 (est.) | 8× Tesla V100 32GBat Vast.ai | $0.54 | $391 | — | Ling 3.0 flash GPU requirements |
| Nemotron Cascade 2 30B A3BNVIDIA · Nemotron-Cascade-2-30B-A3B | 31.6B · 3B active | 63 GBBF16 | 256k | NVIDIA Open Model License | 38 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | Mar 19, 2026 | Nemotron Cascade 2 30B A3B GPU requirements |
| Youtu LLM 2BTencent · Youtu-LLM-2B | 1.96B | 4 GBBF16 | 128k | other | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Youtu LLM 2B GPU requirements |
| Qwen AgentWorld 35B A3BQwen · Qwen-AgentWorld-35B-A3B | 34.7B · 3B active | 69 GBBF16 | 256k | apache-2.0 | 42 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Qwen AgentWorld 35B A3B GPU requirements |
| Solar Open2 250BUpstage · Solar-Open2-250B | 250B · 15B active | 501 GBBF16 | 1M | upstage-solar-license | 331 GBFP8 (est.) | 8× RTX A6000at Lium | $3.36 | $2,453 | — | Solar Open2 250B GPU requirements |
| Ling 2.6 flashinclusionAI · Ling-2.6-flash | 107B · 7.4B active | 215 GBBF16 | 128k | mit | 129 GBFP8 (est.) | 8× Tesla V100 32GBat Vast.ai | $0.54 | $391 | — | Ling 2.6 flash GPU requirements |
| Spark-X2.6-4BXHToken | 4.11B | 8 GBBF16 | 1M | Apache-2.0 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Oct 20, 2026 | Spark-X2.6-4B GPU requirements |
| Muse Spark Open 816B-A42BMeta · Muse-Spark-Open-816B-A42B | 816B · 42B active | 1,644 GBBF16 | 1M | Muse Model License | 980 GBFP8 (est.) | 4× MI325Xat Cyfuture AI | $12.29 | $8,972 | Oct 15, 2026 | Muse Spark Open 816B-A42B GPU requirements |
| Apertus v1.5 70BSwiss AI · Apertus-v1.5-70B | 72B | 145 GBBF16 | — | apache-2.0 | 87 GBFP8 (est.) | 4× RTX 3090at Clore.ai | $0.25 | $183 | — | Apertus v1.5 70B GPU requirements |
| NVIDIA Nemotron-Labs Dory1.0-4BNVIDIA · NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16 | 3.98B | 8 GBBF16 | 256k | OpenMDW-1.1 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Oct 16, 2026 | NVIDIA Nemotron-Labs Dory1.0-4B GPU requirements |
| Nanbeige4.3-33BNanbeige | 33.1B | 66 GBBF16 | 512k | Apache-2.0 | 40 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | Oct 24, 2026 | Nanbeige4.3-33B GPU requirements |
| Beam 1.0 501B-A23BReflection AI · Beam-1.0-501B-A23B | 501B · 23B active | 1,003 GBBF16 | 1M | Apache-2.0 | 602 GBFP8 (est.) | 8× RTX PRO 6000at UpCloud | $7.78 | $5,676 | Oct 16, 2026 | Beam 1.0 501B-A23B GPU requirements |
| LFM2.5-1.2B ThinkingLiquid AI · LFM2.5-1.2B-Thinking | 1.17B | 2 GBBF16 | 125k | other | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | LFM2.5-1.2B Thinking GPU requirements |
| sarvam-105bSarvam | 106B · 10.3B active | 424 GBFP32 | 128k | apache-2.0 | 128 GBFP8 (est.) | 4× Tesla V100 32GBat Vast.ai | $0.27 | $196 | — | sarvam-105b GPU requirements |
| Ming flash omni 2.0inclusionAI · Ming-flash-omni-2.0 | 104B · 6B active | 209 GBBF16 | 32k | mit | 125 GBFP8 (est.) | 4× Tesla V100 32GBat Vast.ai | $0.27 | $196 | — | Ming flash omni 2.0 GPU requirements |
| translategemma-27b-itGoogle | 27.4B | 55 GBBF16 | — | gemma | 33 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | translategemma-27b-it GPU requirements |
| GLM-4 32B 0414Z.ai · GLM-4-32B-0414 | 32.6B | 65 GBBF16 | 32k | mit | 40 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | GLM-4 32B 0414 GPU requirements |
| Llama 3.1 Nemotron Nano 8B v1NVIDIA · Llama-3.1-Nemotron-Nano-8B-v1 | 8.03B | 16 GBBF16 | 128k | other | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Llama 3.1 Nemotron Nano 8B v1 GPU requirements |
| sarvam-30bSarvam | 32.2B · 4.5B active | 129 GBFP32 | 128k | apache-2.0 | 39 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | sarvam-30b GPU requirements |
| MiniCPM4.1 8BOpenBMB · MiniCPM4.1-8B | 8.19B | 16 GBBF16 | 64k | apache-2.0 | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | MiniCPM4.1 8B GPU requirements |
| Ling-3.1-flashinclusionAI | 560B · 25B active | 574 GBFP8 | 256k | MIT | 689 GBFP8 | 8× RTX PRO 6000at UpCloud | $7.78 | $5,676 | Oct 13, 2026 | Ling-3.1-flash GPU requirements |
| NVIDIA NemotronLabs VoiceChat 11BNVIDIA · NVIDIA-NemotronLabs-VoiceChat-11B | 11.1B | 44 GBFP32 | — | openmdw-1.1 | 14 GBFP8 (est.) | 1× T4at Vast.ai | $0.04 | $29 | — | NVIDIA NemotronLabs VoiceChat 11B GPU requirements |
| Ling mini 2.0inclusionAI · Ling-mini-2.0 | 16.3B · 1.4B active | 33 GBBF16 | 32k | mit | 20 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | Ling mini 2.0 GPU requirements |
| Hy-Memory-30B-A3BTencent | 30.06B · 3B active | 60 GBBF16 | 256k | Apache-2.0 | 37 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | Oct 15, 2026 | Hy-Memory-30B-A3B GPU requirements |
| NVIDIA Nemotron 3.5 Bolt E4BNVIDIA · NVIDIA-Nemotron-3.5-Bolt-E4B-BF16 | 8B | 16 GBBF16 | 64k | OpenMDW-1.1 | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Oct 23, 2026 | NVIDIA Nemotron 3.5 Bolt E4B GPU requirements |
| command-a-plus-05-2026-w4a4Cohere | 219B · 25B active | 132 GBINT4-packed | 195k | apache-2.0 | 159 GBINT4-packed | 8× Tesla V100 32GBat Vast.ai | $0.54 | $391 | — | command-a-plus-05-2026-w4a4 GPU requirements |
| Nanbeige4.1 3BNanbeige · Nanbeige4.1-3B | 3.93B | 8 GBBF16 | 256k | apache-2.0 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Nanbeige4.1 3B GPU requirements |
| Phi 4 reasoning plusMicrosoft · Phi-4-reasoning-plus | 14.7B | 29 GBBF16 | 32k | mit | 18 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | Phi 4 reasoning plus GPU requirements |
| Step 5 PreviewStepFun · Step-5-Preview | 600B · 27B active | 650 GBFP8 | 1M | StepFun Model License 1.0 | 781 GBFP8 | 4× MI325Xat Cyfuture AI | $12.29 | $8,972 | Oct 15, 2026 | Step 5 Preview GPU requirements |
| Step Audio 2 miniStepFun · Step-Audio-2-mini | 8.31B | 17 GBBF16 | 16k | apache-2.0 | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Step Audio 2 mini GPU requirements |
| granite-4.0-h-350mIBM | 340M | 1 GBBF16 | 32k | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | granite-4.0-h-350m GPU requirements |
| MiniCPM SALAOpenBMB · MiniCPM-SALA | 9.48B | 19 GBBF16 | 512k | apache-2.0 | 12 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | MiniCPM SALA GPU requirements |
| Granite Switch 4.1 8BIBM · granite-switch-4.1-8b | 8.78B | 18 GBBF16 | 128k | Apache-2.0 | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Oct 14, 2026 | Granite Switch 4.1 8B GPU requirements |
| K-EXAONE 2.0 750B A37BLG AI Research · K-EXAONE-2.0-750B-A37B | 749B · 37B active | 1,499 GBBF16 | 256k | apache-2.0 | 899 GBFP8 (est.) | 4× MI325Xat Cyfuture AI | $12.29 | $8,972 | Jul 31, 2026 | K-EXAONE 2.0 750B A37B GPU requirements |
| Intern S1 ProShanghai AI Lab · Intern-S1-Pro | 917B · 22B active | 919 GBFP8 | 256k | apache-2.0 | 1,103 GBFP8 | 8× MI300Xat TensorWave | $13.68 | $9,986 | — | Intern S1 Pro GPU requirements |
| LFM2-24B A2BLiquid AI · LFM2-24B-A2B | 23.8B · 2B active | 48 GBBF16 | 125k | other | 29 GBFP8 (est.) | 1× Tesla V100 32GBat Vast.ai | $0.07 | $50 | — | LFM2-24B A2B GPU requirements |
| Nex N2.5 miniNex AGI · Nex-N2.5-mini | 35.1B · 3.1B active | 70 GBBF16 | 256k | apache-2.0 | 43 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Nex N2.5 mini GPU requirements |
| DeepSeek V3.2 SpecialeDeepSeek · DeepSeek-V3.2-Speciale | 685B · 37B active | 690 GBFP8 | 160k | mit | 828 GBFP8 | 4× MI325Xat Cyfuture AI | $12.29 | $8,972 | — | DeepSeek V3.2 Speciale GPU requirements |
| granite-4.0-h-1bIBM | 1.46B | 3 GBBF16 | 128k | apache-2.0 | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | granite-4.0-h-1b GPU requirements |
| Phi 4 mini flash reasoningMicrosoft · Phi-4-mini-flash-reasoning | 3.85B | 8 GBBF16 | 256k | mit | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Phi 4 mini flash reasoning GPU requirements |
| Trinity Large ThinkingArcee · Trinity-Large-Thinking | 399B · 13B active | 797 GBBF16 | 256k | OpenMDW-1.1 | 479 GBFP8 (est.) | 2× MI325Xat Cyfuture AI | $6.21 | $4,536 | — | Trinity Large Thinking GPU requirements |
| Raon-1 miniKRAFTON · Raon-1-mini-preview | 28.8B · 4.3B active | 33 GBFP8 + BF16/FP32 | 256k | Apache-2.0 | 40 GBFP8 + BF16/FP32 | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | Oct 21, 2026 | Raon-1 mini GPU requirements |
| LongCat Flash LiteMeituan · LongCat-Flash-Lite | 69.1B | 138 GBBF16 | 320k | mit | 83 GBFP8 (est.) | 4× RTX 3090at Clore.ai | $0.25 | $183 | — | LongCat Flash Lite GPU requirements |
| Mistral Large 3 675B Instruct 2512Mistral AI · Mistral-Large-3-675B-Instruct-2512 | 675B · 41B active | 682 GBFP8 | 288k | apache-2.0 | 818 GBFP8 | 4× MI325Xat Cyfuture AI | $12.29 | $8,972 | — | Mistral Large 3 675B Instruct 2512 GPU requirements |
| Granite Switch 4.1 30BIBM · granite-switch-4.1-30b | 30B | 60 GBBF16 | 128k | Apache-2.0 | 36 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | Oct 14, 2026 | Granite Switch 4.1 30B GPU requirements |
| ZDTaichu5.0 9BTaichu · ZDTaichu5.0-9B | 9.79B | 20 GBBF16 | 256k | — | 12 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | ZDTaichu5.0 9B GPU requirements |
| Motif 3Motif · Motif-3 | 315B | 630 GBBF16 | 256k | mit | 378 GBFP8 (est.) | 8× RTX A6000at Lium | $3.36 | $2,453 | — | Motif 3 GPU requirements |
| Falcon H1R 7BTII · Falcon-H1R-7B | 7.59B | 15 GBBF16 | 256k | other | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Falcon H1R 7B GPU requirements |
| Llama 3.1 Nemotron Ultra 253B v1NVIDIA · Llama-3_1-Nemotron-Ultra-253B-v1 | 253B | 507 GBBF16 | 128k | other | 304 GBFP8 (est.) | 4× A100 80GBat Vast.ai | $2.40 | $1,754 | — | Llama 3.1 Nemotron Ultra 253B v1 GPU requirements |
| Intern S2 PreviewShanghai AI Lab · Intern-S2-Preview | 36.1B · 4.1B active | 73 GBBF16 | 256k | apache-2.0 | 44 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Intern S2 Preview GPU requirements |
| Hunyuan 1.8B InstructTencent · Hunyuan-1.8B-Instruct | 1.79B | 4 GBBF16 | 256k | — | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Hunyuan 1.8B Instruct GPU requirements |
| Nex N2.5 ProNex AGI · Nex-N2.5-Pro | 397B · 11.5B active | 407 GBFP8 | 256k | apache-2.0 | 489 GBFP8 | 2× MI325Xat Cyfuture AI | $6.21 | $4,536 | — | Nex N2.5 Pro GPU requirements |
| Ring 2.5 1TinclusionAI · Ring-2.5-1T | 1T · 63.6B active | 1,015 GBFP8 | 128k | mit | 1,218 GBFP8 | 8× MI300Xat TensorWave | $13.68 | $9,986 | — | Ring 2.5 1T GPU requirements |
| Ling 2.6 1TinclusionAI · Ling-2.6-1T | 1T · 64.3B active | 1,041 GBFP8 | 256k | mit | 1,250 GBFP8 | 8× MI300Xat TensorWave | $13.68 | $9,986 | — | Ling 2.6 1T GPU requirements |
| MiMo Audio 7B InstructXiaomi · MiMo-Audio-7B-Instruct | 8.02B | 16 GBBF16 | 8k | mit | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | MiMo Audio 7B Instruct GPU requirements |
| c4ai-command-a-03-2025Cohere | 111B | 222 GBBF16 | 256k | cc-by-nc-4.0 | 134 GBFP8 (est.) | 8× Tesla V100 32GBat Vast.ai | $0.54 | $391 | — | c4ai-command-a-03-2025 GPU requirements |
| Nanbeige4 3B Thinking 2511Nanbeige · Nanbeige4-3B-Thinking-2511 | 3.93B | 8 GBBF16 | 64k | apache-2.0 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Nanbeige4 3B Thinking 2511 GPU requirements |
| HyperCLOVAX SEED Omni 8BNAVER · HyperCLOVAX-SEED-Omni-8B | 10.7B | 43 GBFP32 | 8k | other | 13 GBFP8 (est.) | 1× T4at Vast.ai | $0.04 | $29 | — | HyperCLOVAX SEED Omni 8B GPU requirements |
| LongCat Flash Thinking 2601Meituan · LongCat-Flash-Thinking-2601 | 562B | 1,124 GBBF16 | 960k | mit | 675 GBFP8 (est.) | 8× RTX PRO 6000at UpCloud | $7.78 | $5,676 | — | LongCat Flash Thinking 2601 GPU requirements |
| Apriel 1.6 15b ThinkerServiceNow · Apriel-1.6-15b-Thinker | 14.9B | 30 GBBF16 | 256k | mit | 18 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | Apriel 1.6 15b Thinker GPU requirements |
| ReaderLM v2Jina AI · ReaderLM-v2 | 1.54B | 3 GBBF16 | 501k | cc-by-nc-4.0 | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | ReaderLM v2 GPU requirements |
| Ring mini 2.0inclusionAI · Ring-mini-2.0 | 16.3B · 1.4B active | 33 GBBF16 | 32k | mit | 20 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | Ring mini 2.0 GPU requirements |
| DeepSeek Math V2DeepSeek · DeepSeek-Math-V2 | 685B · 37B active | 690 GBFP8 | 160k | apache-2.0 | 828 GBFP8 | 4× MI325Xat Cyfuture AI | $12.29 | $8,972 | — | DeepSeek Math V2 GPU requirements |
| EXAONE Deep 32BLG AI Research · EXAONE-Deep-32B | 32B | 64 GBBF16 | 32k | other | 39 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | EXAONE Deep 32B GPU requirements |
| Fun Audio Chat 8BTongyi Lab · Fun-Audio-Chat-8B | 9.45B | 19 GBBF16 | 256k | apache-2.0 | 12 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Fun Audio Chat 8B GPU requirements |
| LongCat NextMeituan · LongCat-Next | 74.3B | 151 GBBF16 | 128k | mit | 90 GBFP8 (est.) | 4× RTX 3090at Clore.ai | $0.25 | $183 | — | LongCat Next GPU requirements |
| LongCat Flash OmniMeituan · LongCat-Flash-Omni | 561B | 1,122 GBBF16 | 128k | mit | 674 GBFP8 (est.) | 8× RTX PRO 6000at UpCloud | $7.78 | $5,676 | — | LongCat Flash Omni GPU requirements |
| Emu3.5BAAI | 34.1B | 68 GBBF16 | 32k | apache-2.0 | 41 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Emu3.5 GPU requirements |
| Atria DawnShanghai AI Lab · Atria-Dawn | 753.3B | 1,507 GBBF16 | 1M | MIT | 904 GBFP8 (est.) | 4× MI325Xat Cyfuture AI | $12.29 | $8,972 | Oct 23, 2026 | Atria Dawn GPU requirements |
| Kimi K3.1 Proweights pendingMoonshot AI · Kimi-K3.1-Pro | 2.8T · 104B active | —BF16 | 1M | — | — | not sizable yet | — | — | Oct 23, 2026 | Kimi K3.1 Pro GPU requirements |
| Mistral Large 4weights pendingMistral AI | 1T · 52B active | —BF16 | — | — | — | not sizable yet | — | — | — | Mistral Large 4 GPU requirements |
Code 16
| Model | Params | Checkpoint | Context | Licence | Needs | Smallest config | $ / hour | $ / month | Release | |
|---|---|---|---|---|---|---|---|---|---|---|
| Qwen3 Coder 30B A3B InstructQwen · Qwen3-Coder-30B-A3B-Instruct | 30.5B · 3B active | 61 GBBF16 | 256k | apache-2.0 | 37 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Qwen3 Coder 30B A3B Instruct GPU requirements |
| Laguna XS.2poolside · Laguna-XS.2 | 33.4B · 3B active | 67 GBBF16 | 256k | apache-2.0 | 41 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Laguna XS.2 GPU requirements |
| Laguna M.1poolside · Laguna-M.1 | 226B · 23B active | 452 GBBF16 | 256k | apache-2.0 | 272 GBFP8 (est.) | 4× A100 80GBat Vast.ai | $2.40 | $1,754 | — | Laguna M.1 GPU requirements |
| Laguna XS 2.1poolside · Laguna-XS-2.1 | 33.4B · 3B active | 67 GBBF16 | 256k | openmdw-1.1 | 41 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Laguna XS 2.1 GPU requirements |
| Laguna-S 2.1Poolside · Laguna-S-2.1 | 118B · 8B active | 235 GBBF16 | 1M | openmdw-1.1 | 156 GBFP8 (est.) | 8× Tesla V100 32GBat Vast.ai | $0.54 | $391 | Jul 21, 2026 | Laguna-S 2.1 GPU requirements |
| Qwen3 Coder NextQwen · Qwen3-Coder-Next | 79.7B · 3B active | 159 GBBF16 | 256k | apache-2.0 | 96 GBFP8 (est.) | 4× RTX 3090at Clore.ai | $0.25 | $183 | — | Qwen3 Coder Next GPU requirements |
| Devstral Small 2 24B Instruct 2512Mistral AI · Devstral-Small-2-24B-Instruct-2512 | 24B | 26 GBFP8 | 384k | apache-2.0 | 31 GBFP8 | 1× Tesla V100 32GBat Vast.ai | $0.07 | $50 | — | Devstral Small 2 24B Instruct 2512 GPU requirements |
| Kimi K2.7 CodeMoonshot AI · Kimi-K2.7-Code | 1T · 32B active | 595 GBINT4-packed | 256k | Modified MIT | 715 GBINT4-packed | 8× RTX PRO 6000at UpCloud | $7.78 | $5,676 | Jun 12, 2026 | Kimi K2.7 Code GPU requirements |
| Qwen3 Coder 480B A35B InstructQwen · Qwen3-Coder-480B-A35B-Instruct | 480B · 35B active | 960 GBBF16 | 256k | apache-2.0 | 576 GBFP8 (est.) | 8× RTX PRO 6000at UpCloud | $7.78 | $5,676 | — | Qwen3 Coder 480B A35B Instruct GPU requirements |
| FlintRecursive · flint | 39.8B · 2.5B active | 80 GBBF16 | 256k | Apache-2.0 | 48 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | Oct 21, 2026 | Flint GPU requirements |
| North Mini Code 1.0Cohere · North-Mini-Code-1.0 | 30.5B · 3B active | 61 GBBF16 | 488k | apache-2.0 | 37 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | North Mini Code 1.0 GPU requirements |
| Mellum2.1-12B-A2.5B-ThinkingJetBrains | 12.15B · 2.5B active | 24 GBBF16 | 128k | Apache-2.0 | 15 GBFP8 (est.) | 1× T4at Vast.ai | $0.04 | $29 | Oct 8, 2026 | Mellum2.1-12B-A2.5B-Thinking GPU requirements |
| KAT Coder V2.5 DevKuaishou · KAT-Coder-V2.5-Dev | 34.7B · 3B active | 69 GBBF16 | 256k | apache-2.0 | 42 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | KAT Coder V2.5 Dev GPU requirements |
| Devstral 2 123B Instruct 2512Mistral AI · Devstral-2-123B-Instruct-2512 | 125B | 128 GBFP8 | 256k | Modified MIT | 154 GBFP8 | 8× Tesla V100 32GBat Vast.ai | $0.54 | $391 | — | Devstral 2 123B Instruct 2512 GPU requirements |
| DiffuCoder 7B cpGRPOApple · DiffuCoder-7B-cpGRPO | 7.62B | 15 GBBF16 | 128k | apple-amlr | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | DiffuCoder 7B cpGRPO GPU requirements |
| Stable DiffCoder 8B InstructByteDance Seed · Stable-DiffCoder-8B-Instruct | 8.25B | 17 GBBF16 | 8k | mit | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Stable DiffCoder 8B Instruct GPU requirements |
Vision 42
Documents and OCR 21
| Model | Params | Checkpoint | Context | Licence | Needs | Smallest config | $ / hour | $ / month | Release | |
|---|---|---|---|---|---|---|---|---|---|---|
| Unlimited OCRBaidu · Unlimited-OCR | 3.34B · 1.1B active | 13 GBBF16 | 32k | mit | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Unlimited OCR GPU requirements |
| GLM OCRZ.ai · GLM-OCR | 1.32B | 3 GBBF16 | 128k | mit | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | GLM OCR GPU requirements |
| chandra-ocr-2Datalab | 5.3B | 11 GBBF16 | 256k | openrail | 7 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | chandra-ocr-2 GPU requirements |
| surya-ocr-2Datalab | 686M | 1 GBBF16 | 256k | openrail | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | surya-ocr-2 GPU requirements |
| HunyuanOCRTencent | 1.12B | 2 GBBF16 | 128k | other | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | HunyuanOCR GPU requirements |
| LightOnOCR 2 1BLightOn · LightOnOCR-2-1B | 1.01B | 2 GBBF16 | 16k | apache-2.0 | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | LightOnOCR 2 1B GPU requirements |
| DeepSeek OCR 2DeepSeek · DeepSeek-OCR-2 | 3.39B · 1.2B active | 7 GBBF16 | 8k | apache-2.0 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | DeepSeek OCR 2 GPU requirements |
| dots.ocrrednote | 3.04B | 6 GBBF16 | 128k | mit | 4 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | dots.ocr GPU requirements |
| Florence 2 largeMicrosoft · Florence-2-large | 777M | 2 GBFP16 | 4k | mit | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Florence 2 large GPU requirements |
| granite docling 258MIBM · granite-docling-258M | 258M | 1 GBBF16 | 8k | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | granite docling 258M GPU requirements |
| dots.mocrrednote | 3.04B | 6 GBBF16 | 128k | mit | 4 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | dots.mocr GPU requirements |
| NVIDIA Nemotron Parse v1.1NVIDIA · NVIDIA-Nemotron-Parse-v1.1 | 957M | 4 GBFP32 | — | other | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | NVIDIA Nemotron Parse v1.1 GPU requirements |
| liftDatalab | 9.65B | 19 GBBF16 | 256k | openrail | 12 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | lift GPU requirements |
| PaddleOCR VL 1.6PaddlePaddle · PaddleOCR-VL-1.6 | 959M | 2 GBBF16 | 128k | apache-2.0 | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | PaddleOCR VL 1.6 GPU requirements |
| MinerU2.5 Pro 2604 1.2BOpenDataLab · MinerU2.5-Pro-2604-1.2B | 1.16B | 2 GBBF16 | 8k | apache-2.0 | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | MinerU2.5 Pro 2604 1.2B GPU requirements |
| Qianfan OCRBaidu · Qianfan-OCR | 4.74B | 10 GBBF16 | 32k | apache-2.0 | 6 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Qianfan OCR GPU requirements |
| NVIDIA Nemotron Parse 2.0NVIDIA · NVIDIA-Nemotron-Parse-2.0 | 903M | 4 GBFP32 | — | openmdw-1.1 | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | NVIDIA Nemotron Parse 2.0 GPU requirements |
| Falcon OCRTII · Falcon-OCR | 270M | 1 GBFP32 | — | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Falcon OCR GPU requirements |
| jina-ocr-v1Jina AI | 3.37B · 1.2B active | 7 GBBF16 | 32k | cc-by-nc-4.0 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | jina-ocr-v1 GPU requirements |
| Dolphin v2ByteDance · Dolphin-v2 | 3.75B | 8 GBBF16 | 125k | — | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Dolphin v2 GPU requirements |
| DolphinByteDance | 398M | 1 GBBF16 | — | mit | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Dolphin GPU requirements |
Image generation 34
| Model | Params | Checkpoint | Context | Licence | Needs | Smallest config | $ / hour | $ / month | Release | |
|---|---|---|---|---|---|---|---|---|---|---|
| stable-diffusion-xl-base-1.0Stability AI | 2.57B | 28 GBFP32 | — | openrail++ | 4 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | stable-diffusion-xl-base-1.0 GPU requirements |
| Z-Image TurboTongyi Lab · Z-Image-Turbo | 6.16B | 33 GBFP32 | — | apache-2.0 | 8 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Nov 25, 2025 | Z-Image Turbo GPU requirements |
| FLUX.1 devBlack Forest Labs · FLUX.1-dev | 11.9B | 24 GBBF16 | — | FLUX.1 [dev] Non-Commercial License | 15 GBFP8 (est.) | 1× T4at Vast.ai | $0.04 | $29 | Jul 31, 2024 | FLUX.1 dev GPU requirements |
| FLUX.1 schnellBlack Forest Labs · FLUX.1-schnell | 11.9B | 24 GBBF16 | — | apache-2.0 | 15 GBFP8 (est.) | 1× T4at Vast.ai | $0.04 | $29 | — | FLUX.1 schnell GPU requirements |
| FLUX.2 devBlack Forest Labs · FLUX.2-dev | 32.2B | 64 GBBF16 | — | FLUX Non-Commercial License | 39 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | Nov 22, 2025 | FLUX.2 dev GPU requirements |
| Qwen Image 2.1Qwen · Qwen-Image-2.1 | 7.12B | 33 GBBF16 | — | Qwen Research License Agreement | 9 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Sep 20, 2026 | Qwen Image 2.1 GPU requirements |
| SDXL LightningByteDance · SDXL-Lightning | 2.6B | 49 GBBF16 | — | openrail++ | 4 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | SDXL Lightning GPU requirements |
| FLUX 3 Image KleinBlack Forest Labs · flux-3-image-klein | 5.6B | 43 GBBF16 | — | FLUX Kommunity License v1.0 | 7 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Oct 15, 2026 | FLUX 3 Image Klein GPU requirements |
| Krea 2 RawKrea · Krea-2-Raw | 12.8B | 62 GBBF16 | — | other | 16 GBFP8 (est.) | 1× T4at Vast.ai | $0.04 | $29 | — | Krea 2 Raw GPU requirements |
| Krea 2 TurboKrea · Krea-2-Turbo | 12.8B | 26 GBBF16 | — | other | 16 GBFP8 (est.) | 1× T4at Vast.ai | $0.04 | $29 | Jun 18, 2026 | Krea 2 Turbo GPU requirements |
| stable-diffusion-3.5-mediumStability AI | 2.47B | 5 GBBF16 | — | other | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | stable-diffusion-3.5-medium GPU requirements |
| Ideogram 4Ideogram · ideogram-4-fp8 | 9.3B | 28 GBFP8 | — | Ideogram 4 Non-Commercial | 12 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Jun 3, 2026 | Ideogram 4 GPU requirements |
| Chroma1 HDLodestones · Chroma1-HD | 8.9B | 45 GBBF16 | — | apache-2.0 | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Chroma1 HD GPU requirements |
| stable-diffusion-3.5-largeStability AI | 8.15B | 16 GBBF16 | — | other | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | stable-diffusion-3.5-large GPU requirements |
| Qwen Image 2512Qwen · Qwen-Image-2512 | 20.4B | 58 GBBF16 | — | apache-2.0 | 25 GBFP8 (est.) | 1× Tesla V100 32GBat Vast.ai | $0.07 | $50 | Dec 30, 2025 | Qwen Image 2512 GPU requirements |
| Z-ImageTongyi Lab | 6.16B | 21 GBBF16 | — | apache-2.0 | 8 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Z-Image GPU requirements |
| Qwen-Image-2.1-TurboQwen | 7.12B | 32 GBBF16 | — | Qwen Research License Agreement | 9 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Oct 9, 2026 | Qwen-Image-2.1-Turbo GPU requirements |
| FLUX.1 Krea devBlack Forest Labs · FLUX.1-Krea-dev | 11.9B | 24 GBBF16 | — | other | 15 GBFP8 (est.) | 1× T4at Vast.ai | $0.04 | $29 | — | FLUX.1 Krea dev GPU requirements |
| MOSS-Image-SFTOpenMOSS | 5.08B | 27 GBBF16 | — | Apache-2.0 | 7 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Oct 16, 2026 | MOSS-Image-SFT GPU requirements |
| LongCat ImageMeituan · LongCat-Image | 6B | 29 GBBF16 | — | apache-2.0 | 8 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | LongCat Image GPU requirements |
| Cosmos3 Super Text2ImageNVIDIA · Cosmos3-Super-Text2Image | 64.6B | 129 GBBF16 | 256k | other | 78 GBFP8 (est.) | 4× RTX 3090at Clore.ai | $0.25 | $183 | — | Cosmos3 Super Text2Image GPU requirements |
| GLM ImageZ.ai · GLM-Image | 6.93B | 36 GBBF16 | — | mit | 9 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | GLM Image GPU requirements |
| HiDream-O1-ImageHiDream | 8.8B | 35 GBFP32 | 256k | MIT | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | May 8, 2026 | HiDream-O1-Image GPU requirements |
| CogView4 6BZ.ai · CogView4-6B | 6.37B | 31 GBBF16 | — | apache-2.0 | 8 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | CogView4 6B GPU requirements |
| SRPOTencent | 11.9B | 48 GBFP32 | — | other | 15 GBFP8 (est.) | 1× T4at Vast.ai | $0.04 | $29 | — | SRPO GPU requirements |
| ERNIE Image TurboBaidu · ERNIE-Image-Turbo | 8.03B | 32 GBBF16 | — | apache-2.0 | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | ERNIE Image Turbo GPU requirements |
| ERNIE ImageBaidu · ERNIE-Image | 8.03B | 32 GBBF16 | — | apache-2.0 | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | ERNIE Image GPU requirements |
| Fibo 2.0 DevBria AI · Fibo-2.0-dev | 8.13B | 26 GBBF16 | — | Bria FIBO licence | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Oct 20, 2026 | Fibo 2.0 Dev GPU requirements |
| Ming Image 0.1 DesigninclusionAI · Ming-Image-0.1-Design | 6.16B | 53 GBBF16 | 32k | mit | 8 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Ming Image 0.1 Design GPU requirements |
| HunyuanImage 3.0Tencent · HunyuanImage-3.0 | 83B | 169 GBBF16 | 22k | Tencent Hunyuan Community License | 100 GBFP8 (est.) | 4× Tesla V100 32GBat Vast.ai | $0.27 | $196 | Sep 28, 2025 | HunyuanImage 3.0 GPU requirements |
| Flex.1 alphaOstris · Flex.1-alpha | 8.16B | 48 GBBF16 | — | apache-2.0 | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Flex.1 alpha GPU requirements |
| HiDream I1 FullHiDream · HiDream-I1-Full | 17.1B | 47 GBFP16 | — | mit | 21 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | HiDream I1 Full GPU requirements |
| Flex.2 previewOstris · Flex.2-preview | 8.16B | 43 GBBF16 | — | apache-2.0 | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Flex.2 preview GPU requirements |
| HunyuanImage 2.1Tencent · HunyuanImage-2.1 | 17B | 166 GBBF16 | — | other | 21 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | HunyuanImage 2.1 GPU requirements |
Image editing 8
| Model | Params | Checkpoint | Context | Licence | Needs | Smallest config | $ / hour | $ / month | Release | |
|---|---|---|---|---|---|---|---|---|---|---|
| FLUX.2 klein 4BBlack Forest Labs · FLUX.2-klein-4B | 3.88B | 8 GBBF16 | — | apache-2.0 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Jan 14, 2026 | FLUX.2 klein 4B GPU requirements |
| FLUX.1 Kontext devBlack Forest Labs · FLUX.1-Kontext-dev | 11.9B | 24 GBBF16 | — | other | 15 GBFP8 (est.) | 1× T4at Vast.ai | $0.04 | $29 | May 28, 2025 | FLUX.1 Kontext dev GPU requirements |
| Qwen Image Edit 2511Qwen · Qwen-Image-Edit-2511 | 20.4B | 58 GBBF16 | — | apache-2.0 | 25 GBFP8 (est.) | 1× Tesla V100 32GBat Vast.ai | $0.07 | $50 | Dec 17, 2025 | Qwen Image Edit 2511 GPU requirements |
| FLUX.2 klein 9BBlack Forest Labs · FLUX.2-klein-9B | 9.08B | 18 GBBF16 | — | FLUX Non-Commercial License | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Jan 14, 2026 | FLUX.2 klein 9B GPU requirements |
| HunyuanImage 3.0 InstructTencent · HunyuanImage-3.0-Instruct | 83B | 169 GBBF16 | 22k | other | 100 GBFP8 (est.) | 4× Tesla V100 32GBat Vast.ai | $0.27 | $196 | — | HunyuanImage 3.0 Instruct GPU requirements |
| LongCat Image EditMeituan · LongCat-Image-Edit | 6B | 29 GBBF16 | — | apache-2.0 | 8 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | LongCat Image Edit GPU requirements |
| Fibo 2.0 Edit TurboBria AI · Fibo-2.0-edit-turbo | 8.13B | 26 GBBF16 | — | Bria FIBO licence | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Oct 20, 2026 | Fibo 2.0 Edit Turbo GPU requirements |
| Step1X EditStepFun · Step1X-Edit | 19B | 50 GBBF16 | — | apache-2.0 | 23 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | Step1X Edit GPU requirements |
Video 23
| Model | Params | Checkpoint | Context | Licence | Needs | Smallest config | $ / hour | $ / month | Release | |
|---|---|---|---|---|---|---|---|---|---|---|
| MiniMax H3MiniMax · MiniMax-H3 | 33.1B | 498 GBBF16 | — | MiniMax H3 Community License Agreement | 40 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | Jul 28, 2026 | MiniMax H3 GPU requirements |
| LTX-2.5Lightricks | 22B | 74 GBBF16 | — | other | 27 GBFP8 (est.) | 1× Tesla V100 32GBat Vast.ai | $0.07 | $50 | Jul 23, 2026 | LTX-2.5 GPU requirements |
| LTX-2Lightricks | 18.9B | 314 GBBF16 | — | other | 23 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | LTX-2 GPU requirements |
| Wan2.1 VACE 14BWan · Wan2.1-VACE-14B | 17.3B | 63 GBFP32 | — | apache-2.0 | 21 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | Wan2.1 VACE 14B GPU requirements |
| Wan2.1 T2V 1.3BWan · Wan2.1-T2V-1.3B | 1.42B | 6 GBFP32 | — | apache-2.0 | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Wan2.1 T2V 1.3B GPU requirements |
| Cosmos3 Super Image2VideoNVIDIA · Cosmos3-Super-Image2Video | 64.6B | 129 GBBF16 | 256k | other | 78 GBFP8 (est.) | 4× RTX 3090at Clore.ai | $0.25 | $183 | — | Cosmos3 Super Image2Video GPU requirements |
| Hy-NomasTencent | 8.55B | 192 GBBF16 | — | Apache-2.0 | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Oct 21, 2026 | Hy-Nomas GPU requirements |
| Wan2.2 S2V 14BWan · Wan2.2-S2V-14B | 16.3B | 33 GBBF16 | — | apache-2.0 | 20 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | Wan2.2 S2V 14B GPU requirements |
| HuMoByteDance | 17B | 58 GBBF16 | — | apache-2.0 | 21 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | HuMo GPU requirements |
| Wan2.2 TI2V 5BWan · Wan2.2-TI2V-5B | 5B | 20 GBBF16 | — | apache-2.0 | 6 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Jul 28, 2025 | Wan2.2 TI2V 5B GPU requirements |
| Wan2.2 I2V A14BWan · Wan2.2-I2V-A14B | 27B · 14B active | 114 GBBF16 | — | apache-2.0 | 33 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Wan2.2 I2V A14B GPU requirements |
| mochi-1-previewGenmo | 10B | 40 GBFP32 | — | apache-2.0 | 12 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | mochi-1-preview GPU requirements |
| Wan2.2 T2V A14BWan · Wan2.2-T2V-A14B | 27B · 14B active | 114 GBBF16 | — | apache-2.0 | 33 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | Jul 28, 2025 | Wan2.2 T2V A14B GPU requirements |
| Wan Dancer 14BWan · Wan-Dancer-14B | 14B | 69 GBBF16 | — | apache-2.0 | 17 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | Wan Dancer 14B GPU requirements |
| stable-video-diffusion-img2vid-xt-1-1Stability AI | 1.52B | 14 GBFP32 | — | other | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | stable-video-diffusion-img2vid-xt-1-1 GPU requirements |
| LongCat VideoMeituan · LongCat-Video | 13.6B | 83 GBBF16 | — | mit | 17 GBFP8 (est.) | 1× Tesla P40at Clore.ai | $0.04 | $30 | Oct 24, 2025 | LongCat Video GPU requirements |
| HunyuanVideo 1.5Tencent · HunyuanVideo-1.5 | 8.3B | 372 GBBF16 | — | Tencent Hunyuan Community License | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Nov 20, 2025 | HunyuanVideo 1.5 GPU requirements |
| Matrix-Game-4.0Skywork AI | 5B | 30 GBBF16 | — | Apache-2.0 | 6 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Oct 12, 2026 | Matrix-Game-4.0 GPU requirements |
| HY WorldPlayTencent · HY-WorldPlay | 8.3B | 147 GBBF16 | — | other | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | HY WorldPlay GPU requirements |
| Matrix Game 2.0Skywork · Matrix-Game-2.0 | 1.8B | 16 GBBF16 | — | mit | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Matrix Game 2.0 GPU requirements |
| Motif Video 2BMotif · Motif-Video-2B | 2.04B | 17 GBFP32 | — | apache-2.0 | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Motif Video 2B GPU requirements |
| Hunyuan GameCraft 1.0Tencent · Hunyuan-GameCraft-1.0 | 13B | 18 GBBF16 | — | — | 23 GBBF16 | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | Hunyuan GameCraft 1.0 GPU requirements |
| HunyuanVideo AvatarTencent · HunyuanVideo-Avatar | 13B | 19 GBBF16 | — | — | 23 GBBF16 | 1× Tesla P40at Clore.ai | $0.04 | $30 | — | HunyuanVideo Avatar GPU requirements |
Voice 34
| Model | Params | Checkpoint | Context | Licence | Needs | Smallest config | $ / hour | $ / month | Release | |
|---|---|---|---|---|---|---|---|---|---|---|
| Kokoro 82MHexgrad · Kokoro-82M | 82M | 0 GBBF16 | — | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Jan 27, 2025 | Kokoro 82M GPU requirements |
| XTTS v2Coqui · XTTS-v2 | 470M | —BF16 | — | other | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | XTTS v2 GPU requirements |
| chatterboxResemble AI | 500M | 11 GBBF16 | — | mit | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | chatterbox GPU requirements |
| Qwen3 TTS 12Hz 1.7B CustomVoiceQwen · Qwen3-TTS-12Hz-1.7B-CustomVoice | 1.92B | 4 GBBF16 | — | apache-2.0 | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Jan 21, 2026 | Qwen3 TTS 12Hz 1.7B CustomVoice GPU requirements |
| F5 TTSF5-TTS · F5-TTS | 340M | 4 GBBF16 | — | cc-by-nc-4.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | F5 TTS GPU requirements |
| VoxCPM2OpenBMB | 2.29B | 5 GBBF16 | — | apache-2.0 | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Apr 3, 2026 | VoxCPM2 GPU requirements |
| Qwen3 TTS 12Hz 1.7B VoiceDesignQwen · Qwen3-TTS-12Hz-1.7B-VoiceDesign | 1.92B | 4 GBBF16 | — | apache-2.0 | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Qwen3 TTS 12Hz 1.7B VoiceDesign GPU requirements |
| VibeVoice 1.5BMicrosoft · VibeVoice-1.5B | 2.7B | 5 GBBF16 | — | mit | 4 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Aug 25, 2025 | VibeVoice 1.5B GPU requirements |
| Fun CosyVoice3 0.5B 2512Tongyi Lab · Fun-CosyVoice3-0.5B-2512 | 500M | 1 GBBF16 | — | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Fun CosyVoice3 0.5B 2512 GPU requirements |
| csm-1bSesame | 1.55B | 13 GBFP32 | 2k | apache-2.0 | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | csm-1b GPU requirements |
| higgs-tts-3-4bBoson AI | 4.66B | 9 GBBF16 | 32k | other | 6 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Jun 4, 2026 | higgs-tts-3-4b GPU requirements |
| MOSS TTS Nano 100MOpenMOSS · MOSS-TTS-Nano-100M | 100M | 0 GBBF16 | — | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | MOSS TTS Nano 100M GPU requirements |
| MOSS TTS v1.5OpenMOSS · MOSS-TTS-v1.5 | 8.49B | 17 GBBF16 | 40k | apache-2.0 | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | MOSS TTS v1.5 GPU requirements |
| supertonic-3Supertone | 70M | —BF16 | — | openrail | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | supertonic-3 GPU requirements |
| KittenTTS 2KittenML · kitten-tts-2 | 1.7B | 5 GBBF16 | — | Stellon Labs Community License | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Sep 30, 2026 | KittenTTS 2 GPU requirements |
| tts-1.6b-en_frKyutai | 1.6B | 4 GBBF16 | — | cc-by-4.0 | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | tts-1.6b-en_fr GPU requirements |
| s2-proFish Audio | 4.56B | 9 GBBF16 | — | Fish Audio Research License | 6 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Mar 9, 2026 | s2-pro GPU requirements |
| Fish Audio Drama 3Fish Audio · drama-3 | 35B | 42 GBFP8 + BF16 | 256k | Fish Audio Community License v1.0 | 50 GBFP8 + BF16 | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | Oct 13, 2026 | Fish Audio Drama 3 GPU requirements |
| orpheus-3b-0.1-ftCanopy Labs | 3.78B | 15 GBFP32 | 128k | apache-2.0 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | orpheus-3b-0.1-ft GPU requirements |
| IndexTTS 2.5Bilibili · IndexTTS-2.5 | 1.5B | 1 GBBF16 | — | other | 2 GBBF16 | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | IndexTTS 2.5 GPU requirements |
| Breeze TTS 2BreezeBlue · Breeze-TTS-2 | 3.48B | 8 GBBF16 | 2k | BreezeBlue Research and Non-Commercial L | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Aug 25, 2026 | Breeze TTS 2 GPU requirements |
| magpie_tts_multilingual_357mNVIDIA | 360M | —BF16 | — | other | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | magpie_tts_multilingual_357m GPU requirements |
| Kimi Audio 7B InstructMoonshot AI · Kimi-Audio-7B-Instruct | 9.77B | 20 GBBF16 | 8k | mit | 12 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Kimi Audio 7B Instruct GPU requirements |
| Dia2 2BNari Labs · Dia2-2B | 1.92B | 8 GBFP32 | — | apache-2.0 | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Dia2 2B GPU requirements |
| VoxCPM 0.5BOpenBMB · VoxCPM-0.5B | 500M | 1 GBBF16 | — | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | VoxCPM 0.5B GPU requirements |
| Ming-3.0-tts-0.5BinclusionAI | 500M | 3 GBBF16 | 32k | MIT | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Sep 22, 2026 | Ming-3.0-tts-0.5B GPU requirements |
| VoxCPM1.5OpenBMB | 802M | 2 GBBF16 | — | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | VoxCPM1.5 GPU requirements |
| s1-miniFish Audio | 500M | —BF16 | — | cc-by-nc-sa-4.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | s1-mini GPU requirements |
| Voxtral 4B TTS 2603Mistral AI · Voxtral-4B-TTS-2603 | 4B | 8 GBBF16 | — | cc-by-nc-4.0 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Voxtral 4B TTS 2603 GPU requirements |
| neutts-airNeuphonic | 748M | 2 GBBF16 | — | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | neutts-air GPU requirements |
| Zonos v0.1 hybridZyphra · Zonos-v0.1-hybrid | 1.65B | 3 GBBF16 | — | apache-2.0 | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Zonos v0.1 hybrid GPU requirements |
| GLM TTSZ.ai · GLM-TTS | 1.5B | 8 GBBF16 | — | mit | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | GLM TTS GPU requirements |
| MegaTTS3ByteDance | 450M | 1 GBBF16 | — | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | MegaTTS3 GPU requirements |
| chatterbox-turboResemble AI | 350M | 4 GBBF16 | — | mit | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Dec 2, 2025 | chatterbox-turbo GPU requirements |
Speech to text 23
| Model | Params | Checkpoint | Context | Licence | Needs | Smallest config | $ / hour | $ / month | Release | |
|---|---|---|---|---|---|---|---|---|---|---|
| whisper-large-v3-turboOpenAI | 809M | 2 GBFP16 | — | mit | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Oct 1, 2024 | whisper-large-v3-turbo GPU requirements |
| whisper-large-v3OpenAI | 1.54B | 3 GBFP16 | — | apache-2.0 | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Nov 7, 2023 | whisper-large-v3 GPU requirements |
| Voxtral Mini 4B Realtime 2602Mistral AI · Voxtral-Mini-4B-Realtime-2602 | 4.43B | 9 GBBF16 | 128k | apache-2.0 | 6 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Jan 21, 2026 | Voxtral Mini 4B Realtime 2602 GPU requirements |
| Qwen3 ASR 1.7BQwen · Qwen3-ASR-1.7B | 2.35B | 5 GBBF16 | — | apache-2.0 | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Jan 28, 2026 | Qwen3 ASR 1.7B GPU requirements |
| nemotron-3.5-asr-streaming-0.6bNVIDIA | 638M | 3 GBFP32 | — | OpenMDW-1.1 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | May 15, 2026 | nemotron-3.5-asr-streaming-0.6b GPU requirements |
| Qwen3 ASR 0.6BQwen · Qwen3-ASR-0.6B | 938M | 2 GBBF16 | — | apache-2.0 | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Qwen3 ASR 0.6B GPU requirements |
| VibeVoice ASRMicrosoft · VibeVoice-ASR | 8.67B | 17 GBBF16 | — | mit | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Jan 21, 2026 | VibeVoice ASR GPU requirements |
| parakeet-tdt-0.6b-v3NVIDIA | 627M | 3 GBFP32 | — | cc-by-4.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Aug 4, 2025 | parakeet-tdt-0.6b-v3 GPU requirements |
| seamless-m4t-v2-largeMeta | 2.31B | 9 GBFP32 | 4k | cc-by-nc-4.0 | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | seamless-m4t-v2-large GPU requirements |
| granite-4.0-1b-speechIBM | 2.31B | 5 GBBF16 | 4k | apache-2.0 | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | granite-4.0-1b-speech GPU requirements |
| canary-1b-v2NVIDIA | 979M | 4 GBFP32 | — | cc-by-4.0 | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Aug 4, 2025 | canary-1b-v2 GPU requirements |
| granite-speech-4.1-2bIBM | 2.31B | 5 GBBF16 | 4k | apache-2.0 | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | granite-speech-4.1-2b GPU requirements |
| cohere-transcribe-arabic-07-2026Cohere | 2.07B | 4 GBBF16 | — | apache-2.0 | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | cohere-transcribe-arabic-07-2026 GPU requirements |
| parakeet-tdt-0.6b-v2NVIDIA | 600M | —BF16 | — | cc-by-4.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | parakeet-tdt-0.6b-v2 GPU requirements |
| nemotron-speech-streaming-en-0.6bNVIDIA | 618M | 3 GBFP32 | — | other | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | nemotron-speech-streaming-en-0.6b GPU requirements |
| Phi 4 multimodal instructMicrosoft · Phi-4-multimodal-instruct | 5.57B | 11 GBBF16 | 128k | mit | 7 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | Phi 4 multimodal instruct GPU requirements |
| cohere-transcribe-03-2026Cohere | 2.07B | 4 GBBF16 | — | apache-2.0 | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Mar 24, 2026 | cohere-transcribe-03-2026 GPU requirements |
| GLM ASR Nano 2512Z.ai · GLM-ASR-Nano-2512 | 2.26B | 5 GBBF16 | 8k | mit | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | GLM ASR Nano 2512 GPU requirements |
| Granite Speech 5.0 2BIBM · granite-speech-5.0-2b | 2.31B | 5 GBBF16 | 16k | Apache-2.0 | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Oct 7, 2026 | Granite Speech 5.0 2B GPU requirements |
| Fun ASR Nano 2512Tongyi Lab · Fun-ASR-Nano-2512 | 800M | —BF16 | — | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Fun ASR Nano 2512 GPU requirements |
| Voxtral Mini 4B Realtime ArabicMistral AI · Voxtral-Mini-4B-Realtime-Arabic | 4.43B | 9 GBBF16 | 128k | Apache-2.0 | 6 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Oct 8, 2026 | Voxtral Mini 4B Realtime Arabic GPU requirements |
| VibeVoice ASR Streaming 7BMicrosoft · VibeVoice-ASR-Streaming-7B | 8.67B | 17 GBBF16 | — | mit | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | VibeVoice ASR Streaming 7B GPU requirements |
| Ming-3.0-asr-tinyinclusionAI | 8.87B | 18 GBBF16 | 8k | MIT | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Oct 14, 2026 | Ming-3.0-asr-tiny GPU requirements |
Music 10
| Model | Params | Checkpoint | Context | Licence | Needs | Smallest config | $ / hour | $ / month | Release | |
|---|---|---|---|---|---|---|---|---|---|---|
| Ace Step1.5ACE Studio · Ace-Step1.5 | 3.5B | 10 GBBF16 | 32k | mit | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Jan 23, 2026 | Ace Step1.5 GPU requirements |
| stable-audio-3-mediumStability AI | 2.31B | 9 GBFP32 | — | other | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | May 17, 2026 | stable-audio-3-medium GPU requirements |
| musicgen-largeMeta | 3.3B | 14 GBBF16 | — | cc-by-nc-4.0 | 4 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | musicgen-large GPU requirements |
| YuE2 3BM-A-P · YuE2-3B | 3.63B | 7 GBBF16 | 24k | cc-by-nc-4.0 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Sep 9, 2026 | YuE2 3B GPU requirements |
| stable-audio-open-1.0Stability AI | 1.21B | 5 GBFP32 | — | other | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | stable-audio-open-1.0 GPU requirements |
| acestep-v15-xl-turboACE Studio | 4.99B | 20 GBFP32 | 32k | mit | 6 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Apr 2, 2026 | acestep-v15-xl-turbo GPU requirements |
| stable-audio-open-smallStability AI | 497M | 2 GBFP32 + FP16 | — | other | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | stable-audio-open-small GPU requirements |
| MiniMax Music3MiniMax · MiniMax-Music3 | 11B | 47 GBFP32 | — | — | 14 GBFP8 (est.) | 1× T4at Vast.ai | $0.04 | $29 | Aug 7, 2026 | MiniMax Music3 GPU requirements |
| YuE s1 7B anneal en cotM-A-P · YuE-s1-7B-anneal-en-cot | 6.22B | 12 GBBF16 | 16k | apache-2.0 | 8 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | YuE s1 7B anneal en cot GPU requirements |
| HeartMuLa-oss-3BHeartMuLa | 3.94B | 16 GBFP32 | — | Apache-2.0 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Jan 14, 2026 | HeartMuLa-oss-3B GPU requirements |
3D 14
| Model | Params | Checkpoint | Context | Licence | Needs | Smallest config | $ / hour | $ / month | Release | |
|---|---|---|---|---|---|---|---|---|---|---|
| TRELLIS.2-4BMicrosoft | 4B | 16 GBBF16 | — | mit | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Dec 1, 2025 | TRELLIS.2-4B GPU requirements |
| Hunyuan3D 2Tencent · Hunyuan3D-2 | 2.6B | 34 GBBF16 | — | other | 4 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Hunyuan3D 2 GPU requirements |
| Hunyuan3D 2.1Tencent · Hunyuan3D-2.1 | 5.3B | 1 GBBF16 | — | Tencent Hunyuan Community License | 2 GBBF16 | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Jun 13, 2025 | Hunyuan3D 2.1 GPU requirements |
| stable-fast-3dStability AI | 1.01B | 4 GBFP32 | — | other | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | stable-fast-3d GPU requirements |
| stable-virtual-cameraStability AI | 1.26B | 10 GBFP32 | — | other | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | stable-virtual-camera GPU requirements |
| Hunyuan3D 2mvTencent · Hunyuan3D-2mv | 1.1B | 15 GBBF16 | — | other | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Hunyuan3D 2mv GPU requirements |
| TripoSGVAST AI | 1.5B | 8 GBFP32 | — | MIT | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Mar 24, 2025 | TripoSG GPU requirements |
| SAM 3D ObjectsMeta · sam-3d-objects | — | 13 GB | — | SAM License | 16 GBBF16 | 1× T4at Vast.ai | $0.04 | $29 | Nov 19, 2025 | SAM 3D Objects GPU requirements |
| Hunyuan3D OmniTencent · Hunyuan3D-Omni | 3.3B | —BF16 | — | other | 4 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Hunyuan3D Omni GPU requirements |
| HunyuanWorld MirrorTencent · HunyuanWorld-Mirror | 1.26B | 5 GBFP32 | — | other | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | HunyuanWorld Mirror GPU requirements |
| HY Motion 1.0Tencent · HY-Motion-1.0 | 1B | —BF16 | — | other | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | HY Motion 1.0 GPU requirements |
| Step1X-3DStepFun | 4.8B | 20 GBBF16 | — | Apache-2.0 | 6 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | May 13, 2025 | Step1X-3D GPU requirements |
| Hi3D v1.0Hi3D · Hi3Dv1.0 | — | 8 GB | — | Hi3D Open License Agreement 1.0 | 10 GBBF16 | 1× RTX 3060at Vast.ai | $0.04 | $26 | Oct 23, 2026 | Hi3D v1.0 GPU requirements |
| Pixal3DTencent | — | 46 GB | — | MIT | 56 GBBF16 | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | — | Pixal3D GPU requirements |
Embeddings and rerank 43
| Model | Params | Checkpoint | Context | Licence | Needs | Smallest config | $ / hour | $ / month | Release | |
|---|---|---|---|---|---|---|---|---|---|---|
| all MiniLM L6 v2Sentence Transformers · all-MiniLM-L6-v2 | 23M | 0 GBFP32 | 1k | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | all MiniLM L6 v2 GPU requirements |
| paraphrase multilingual MiniLM L12 v2Sentence Transformers · paraphrase-multilingual-MiniLM-L12-v2 | 118M | 0 GBFP32 | 1k | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | paraphrase multilingual MiniLM L12 v2 GPU requirements |
| bge-m3BAAI | 570M | 2 GBBF16 | 8k | mit | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Jan 27, 2024 | bge-m3 GPU requirements |
| all-mpnet-base-v2Sentence Transformers | 109M | 0 GBFP32 | 1k | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | all-mpnet-base-v2 GPU requirements |
| bge-reranker-v2-m3BAAI | 568M | 2 GBFP32 | 8k | Apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Mar 15, 2024 | bge-reranker-v2-m3 GPU requirements |
| nomic-embed-text-v1.5Nomic AI | 137M | 1 GBFP32 | 2k | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | nomic-embed-text-v1.5 GPU requirements |
| Qwen3 Embedding 0.6BQwen · Qwen3-Embedding-0.6B | 596M | 1 GBBF16 | 32k | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Jun 3, 2025 | Qwen3 Embedding 0.6B GPU requirements |
| multilingual-e5-largeMicrosoft | 560M | 2 GBFP32 | 1k | mit | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | multilingual-e5-large GPU requirements |
| Qwen3-Reranker-4BQwen | 4.02B | 8 GBBF16 | 40k | Apache-2.0 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Jun 3, 2025 | Qwen3-Reranker-4B GPU requirements |
| mxbai-embed-large-v1Mixedbread | 335M | 1 GBFP16 | 1k | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | mxbai-embed-large-v1 GPU requirements |
| Qwen3 Embedding 8BQwen · Qwen3-Embedding-8B | 7.57B | 15 GBBF16 | 40k | apache-2.0 | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Jun 3, 2025 | Qwen3 Embedding 8B GPU requirements |
| Qwen3 Embedding 4BQwen · Qwen3-Embedding-4B | 4.02B | 8 GBBF16 | 40k | apache-2.0 | 5 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Jun 3, 2025 | Qwen3 Embedding 4B GPU requirements |
| multilingual-e5-large-instructMicrosoft | 560M | 1 GBFP16 | 1k | mit | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | multilingual-e5-large-instruct GPU requirements |
| embeddinggemma-300mGoogle | 303M | 1 GBFP32 | 2k | gemma | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | embeddinggemma-300m GPU requirements |
| jina-embeddings-v3Jina AI | 572M | 1 GBBF16 | 8k | cc-by-nc-4.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | jina-embeddings-v3 GPU requirements |
| Qwen3 Reranker 0.6BQwen · Qwen3-Reranker-0.6B | 596M | 1 GBBF16 | 40k | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | May 29, 2025 | Qwen3 Reranker 0.6B GPU requirements |
| Qwen3 VL Embedding 8BQwen · Qwen3-VL-Embedding-8B | 8.14B | 16 GBBF16 | 256k | apache-2.0 | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Jan 7, 2026 | Qwen3 VL Embedding 8B GPU requirements |
| jina-embeddings-v5-omni-smallJina AI | 1.63B | 3 GBBF16 | 32k | cc-by-nc-4.0 | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Apr 1, 2026 | jina-embeddings-v5-omni-small GPU requirements |
| nomic-embed-text-v2-moeNomic AI | 475M | 2 GBFP32 | — | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Feb 7, 2025 | nomic-embed-text-v2-moe GPU requirements |
| Qwen3 VL Embedding 2BQwen · Qwen3-VL-Embedding-2B | 2.13B | 4 GBBF16 | 256k | apache-2.0 | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Qwen3 VL Embedding 2B GPU requirements |
| EmbeddingGemma 2Google · embeddinggemma-2 | 740M | 2 GBBF16 | 256k | Apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Oct 6, 2026 | EmbeddingGemma 2 GPU requirements |
| harrier-oss-v1-0.6bMicrosoft | 596M | 1 GBBF16 | 32k | mit | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | harrier-oss-v1-0.6b GPU requirements |
| Nemotron-3-Embed-8BNVIDIA · Nemotron-3-Embed-8B-BF16 | 7.95B | 16 GBBF16 | 256k | OpenMDW-1.1 | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Jul 14, 2026 | Nemotron-3-Embed-8B GPU requirements |
| jina-reranker-v3.5Jina AI | 597M | 1 GBBF16 | 128k | CC BY-NC 4.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Jul 14, 2026 | jina-reranker-v3.5 GPU requirements |
| granite-embedding-97m-multilingual-r2IBM | 97M | 0 GBBF16 | 32k | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | granite-embedding-97m-multilingual-r2 GPU requirements |
| harrier-oss-v1-270mMicrosoft | 268M | 1 GBBF16 | 32k | mit | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | harrier-oss-v1-270m GPU requirements |
| granite-embedding-311m-multilingual-r2IBM | 312M | 1 GBBF16 | 32k | apache-2.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | granite-embedding-311m-multilingual-r2 GPU requirements |
| Qwen3 Reranker 8BQwen · Qwen3-Reranker-8B | 8.19B | 16 GBBF16 | 40k | apache-2.0 | 10 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | May 29, 2025 | Qwen3 Reranker 8B GPU requirements |
| Qwen3 VL Reranker 2BQwen · Qwen3-VL-Reranker-2B | 2.13B | 4 GBBF16 | 256k | apache-2.0 | 3 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Qwen3 VL Reranker 2B GPU requirements |
| Reason ModernColBERTLightOn · Reason-ModernColBERT | 149M | 1 GBFP32 | 8k | cc-by-nc-4.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | Reason ModernColBERT GPU requirements |
| Qwen3 VL Reranker 8BQwen · Qwen3-VL-Reranker-8B | 8.77B | 18 GBBF16 | 256k | apache-2.0 | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Jan 7, 2026 | Qwen3 VL Reranker 8B GPU requirements |
| LFM2.5 Embedding 350MLiquid AI · LFM2.5-Embedding-350M | 354M | 1 GBBF16 | 125k | other | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | LFM2.5 Embedding 350M GPU requirements |
| llama-embed-nemotron-8bNVIDIA | 7.5B | 15 GBBF16 | 128k | other | 9 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | llama-embed-nemotron-8b GPU requirements |
| LFM2.5 ColBERT 350MLiquid AI · LFM2.5-ColBERT-350M | 353M | 1 GBBF16 | 125k | other | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | LFM2.5 ColBERT 350M GPU requirements |
| SheetSage2M-A-P | 57M | 0 GBFP32 | — | cc-by-nc-4.0 | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | SheetSage2 GPU requirements |
| Nemotron-3.5-Embed-1BNVIDIA · Nemotron-3.5-Embed-1B-BF16 | 1.55B | 3 GBBF16 | 256k | OpenMDW-1.1 | 2 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | Oct 20, 2026 | Nemotron-3.5-Embed-1B GPU requirements |
| harrier-oss-v1-27bMicrosoft | 27B | 54 GBBF16 | 128k | mit | 33 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | Mar 30, 2026 | harrier-oss-v1-27b GPU requirements |
| Nemotron-3.5-Embed-8BNVIDIA · Nemotron-3.5-Embed-8B-BF16 | 8.38B | 17 GBBF16 | 256k | OpenMDW-1.1 | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Oct 20, 2026 | Nemotron-3.5-Embed-8B GPU requirements |
| LFM2 ColBERT 350MLiquid AI · LFM2-ColBERT-350M | 353M | 1 GBFP32 | 125k | other | 1 GBFP8 (est.) | 1× Quadro P2000at Vast.ai | $0.03 | $22 | — | LFM2 ColBERT 350M GPU requirements |
| Nemotron-3.5-Rerank-8BNVIDIA · Nemotron-3.5-Rerank-8B-BF16 | 8.38B | 17 GBBF16 | 256k | OpenMDW-1.1 | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Oct 20, 2026 | Nemotron-3.5-Rerank-8B GPU requirements |
| omni-embed-nemotron-3bNVIDIA | 4.7B | 9 GBBF16 | 32k | other | 6 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | — | omni-embed-nemotron-3b GPU requirements |
| pplx-embed-v2-context-9b-previewPerplexity | 8.4B | 34 GBFP32 | 256k | MIT | 11 GBFP8 (est.) | 1× RTX 3060at Vast.ai | $0.04 | $26 | Sep 25, 2026 | pplx-embed-v2-context-9b-preview GPU requirements |
| MetaEncoder-30BMeta · meta-encoder | 29.78B | 60 GBBF16 | — | Apache-2.0 | 36 GBFP8 (est.) | 2× Tesla V100 32GBat Vast.ai | $0.14 | $99 | Sep 30, 2026 | MetaEncoder-30B GPU requirements |
Needs = weights at the precision plus 20% runtime overhead, short context · smallest config = fewest GPUs of one type whose total VRAM fits, among 1, 2, 4 and 8 · prices from the cheapest fresh on-demand offer for that count · 412 models carry the study's throughput estimatesHow sizing works →