Mistral AI · Chat and reasoning · 128B parameters · Modified MIT
Mistral Medium 3.5 128B
Needs 161 GB at FP8 / INT8. The smallest real configuration today is 8× Tesla V100 32GB at Vast.ai, $0.54 per hour, about $391 a month at 730 hours.
Parameters128Bdense
Context256k tokensmaximum published
Checkpoint134 GBFP8, as published
LicenseModified MITcheck before commercial use
ReleasedApr 29, 2026documented date
Smallest configuration · FP8 / INT8
8× Tesla V100 32GB256 GB VRAM · 63% used
- Per hour
- $0.54
- Per month
- $391
- Cheapest at
Vast.ai
Memory · 2 precisions · 5 context lengthsHow much VRAM it needs
Weights plus runtime overhead, then the KV cache that grows with the context you serve.
| Precision | Weights | Weights + headroom | 4k context | 32k context | 128k context | 256k context |
|---|---|---|---|---|---|---|
| FP8published checkpoint | 134 GB | 161 GB | 162 GB | 173 GB | 208 GB | 255 GB |
| INT4 / MXFP4estimated | 77 GB | 93 GB | 94 GB | 104 GB | 140 GB | 187 GB |
| KV cacheone sequence, FP16 cache, included above | — | — | 2 GB | 12 GB | 48 GB | 95 GB |
Cross-checked on Hugging Face Oct 28, 2026: 127.7B parameters in the safetensors headers, 133.6 GB of weight files, attention layout from config.json (88 layers, 8 KV heads × 128). Method: weights as published at the native precision, verified on Hugging Face 2026-10-28; other precisions estimated at 1 (8-bit) and 0.6 (4-bit) bytes per parameter; × 1.2 runtime headroom; KV cache = 4096 bytes per token per layer × 88 full-attention layers (FP16 cache, one sequence).config.json ↗
Smallest configuration per GPU · 109 GPUs · cheapest eligible on-demand offer per GPU count, every provider on fileWhat it costs to run, GPU by GPU
The fewest GPUs of one type whose total memory fits the model at short context, priced from the cheapest eligible on-demand offer for exactly that count (45-day window, newest reading per provider), with the next providers beside it. Monthly = 730 hours.
| GPU | Smallest config | VRAM used | $ / hour | $ / month | Cheapest at |
|---|---|---|---|---|---|
| Tesla V100 32GBBest fit | 8× 32 GB256 GB total | 63% | $0.54 | $391 | Vast.aiTrust 78 · read Nov 2 · receipt · also Clore.ai $0.69, Edgevana $2.21 |
| Quadro RTX 8000 | 4× 48 GB192 GB total | 84% | $1.01 | $741 | Vast.aiTrust 78 · read Nov 2 · receipt |
| RTX 3090 | 8× 24 GB192 GB total | 84% | $1.17 | $851 |  Clore.aiTrust 49 · read Nov 2 · receipt · also Vast.ai $1.29, Lium $2.32 |
| L40S | 4× 48 GB192 GB total | 84% | $1.52 | $1,110 |  LiumTrust 40 · read Nov 2 · receipt · also Vast.ai $1.87, VoltageGPU $3.04 |
| MI300X | 1× 192 GB192 GB total | 84% | $1.65 | $1,205 |  GPUBankTrust 93 · read Oct 25 · receipt · also OpenRelay $1.99, Runpod $2.39 |
| RTX PRO 6000 | 2× 96 GB192 GB total | 84% | $1.67 | $1,218 | Vast.aiTrust 78 · read Nov 2 · receipt · also UpCloud $1.94, GPU.ai $3.38 |
| RTX A6000 | 4× 48 GB192 GB total | 84% | $1.68 | $1,226 | LiumTrust 40 · read Nov 2 · receipt · also Spheron $1.77, Vast.ai $1.82 |
| RTX PRO 5000 | 4× 48 GB192 GB total | 84% | $1.92 | $1,403 | Vast.aiTrust 78 · read Nov 2 · receipt · also Edgevana $4.60, iwinv $6.66 |
| RTX PRO 4000 | 8× 24 GB192 GB total | 84% | $1.93 | $1,406 | Vast.aiTrust 78 · read Nov 2 · receipt · also Edgevana $2.15, AtmosCompute $3.37 |
| H100 NVL | 2× 94 GB188 GB total | 85% | $2.08 | $1,521 | Clore.aiTrust 49 · read Nov 2 · receipt · also Vast.ai $5.23, AtmosCompute $8.93 |
| Gaudi 2 | 2× 96 GB192 GB total | 84% | $2.11 | $1,538 |  Cyfuture AInot scored · read Nov 2 · receipt |
| MI325X | 1× 256 GB256 GB total | 63% | $2.25 | $1,643 |  OpenRelaynot scored · read Nov 2 · receipt · also Cyfuture AI $3.14, DigitalOcean $3.80 |
| A40 | 4× 48 GB192 GB total | 84% | $2.30 | $1,676 | Vast.aiTrust 78 · read Nov 2 · receipt · also Denvr Cloud $2.60, Edgevana $3.01 |
| A100 80GB | 4× 80 GB320 GB total | 50% | $2.40 | $1,754 | Vast.aiTrust 78 · read Nov 2 · receipt · also Edgevana $4.60, Gcore $5.63 |
| RTX 6000 Ada | 4× 48 GB192 GB total | 84% | $2.60 | $1,898 |  GPU.ainot scored · read Nov 2 · receipt · also io.net $2.73, QuantaCloud $3.11 |
| L40 | 4× 48 GB192 GB total | 84% | $2.65 | $1,932 |  io.netTrust 42 · read Nov 2 · receipt · also GPU.ai $3.12, QuantaCloud $3.75 |
| RTX 4090 | 8× 24 GB192 GB total | 84% | $3.07 | $2,241 |  Edgevananot scored · read Nov 2 · receipt · also Sesterce Cloud $3.52, Vast.ai $3.63 |
| A100 40GB | 8× 40 GB320 GB total | 50% | $3.20 | $2,338 | Vast.aiTrust 78 · read Nov 2 · receipt · also Denvr Cloud $9.20, Sesterce Cloud $12.32 |
| B200 | 1× 180 GB180 GB total | 89% | $3.75 | $2,738 |  packet.ainot scored · read Nov 2 · receipt · also Beam $4.11, UpCloud $4.11 |
| RTX 5090 | 8× 32 GB256 GB total | 63% | $3.84 | $2,804 | Vast.aiTrust 78 · read Nov 2 · receipt · also Lium $4.00, Edgevana $5.28 |
| MI350X | 1× 288 GB288 GB total | 56% | $3.99 | $2,913 | OpenRelaynot scored · read Nov 2 · receipt |
| A100X | 4× 80 GB320 GB total | 50% | $5.12 | $3,739 | Vast.aiTrust 78 · read Nov 2 · receipt |
| H200 NVL | 2× 141 GB282 GB total | 57% | $5.33 | $3,894 | Vast.aiTrust 78 · read Nov 2 · receipt · also IONOS $6.52, Geodd $6.78 |
| A30 | 8× 24 GB192 GB total | 84% | $5.42 | $3,960 |  Leafcloudnot scored · read Nov 2 · receipt |
| The 24 cheapest of 36 GPU types that fit. Every GPU, any precision and context, in the finder | |||||
Throughput estimates · Oct 4, 2026Tokens per second, and what a million costs
Chat and reasoning. Speed estimated from memory bandwidth and model size, never measured. The cost per 1M output tokens is an estimate at the prices read on Oct 4, 2026; it is not recomputed here.
| GPU | Cards | Precision | tok/s per user | per 1M output tokens · estimate |
|---|---|---|---|---|
| MI300X | 1 | FP8 | 20.7 | $0.963 |
| H200 | 2 | FP8 | 37.3 | $1.46 |
| B200 | 1 | FP8 | 38.1 | $1.48 |
| RTX PRO 6000 | 2 | FP8 | 13.4 | $1.97 |
| L40S | 4 | FP8 | 10.6 | $2.19 |
| H100 SXM | 2 | FP8 | 27.9 | $5.28 |
| A100 80GB | 2 | FP8 | 17.8 | $5.69 |
| H100 PCIe | 2 | FP8 | 14.5 | $7.22 |
cards, precision and speed estimated from memory bandwidth and model size; cost recomputed by the site from verified prices · cheapest per output at the prices of Oct 4, 2026: MI300XWhy estimates, and their limits →
Mistral Medium 3.5 128B · related