DeepSeek · Chat and reasoning · 763.2B parameters · 16B active · mit

DeepSeek V4.1 Flash

Needs 613 GB at INT4 / MXFP4. The smallest real configuration today is 8× RTX PRO 6000 at UpCloud, $7.78 per hour, about $5,676 a month at 730 hours.

Parameters763.2B16B active per token

Context1M tokensmaximum published

Checkpoint510 GBINT4-packed + FP8/BF16, as published

Licensemitcheck before commercial use

ReleasedSep 10, 2026updated Oct 17, 2026

Downloads116,770deepseek-ai/DeepSeek-V4.1-Flash-1017 ↗

Smallest configuration · INT4 / MXFP4

8× RTX PRO 6000768 GB VRAM · 80% used

Per hour: $7.78
Per month: $5,676
Cheapest at: UpCloud

All RTX PRO 6000 offers →

What is documented about DeepSeek V4.1 Flash

news checked 2026-10-27

Causal encoder-decoder MoE released 2026-09-10 with weights under MIT. The two figures in circulation are both right: a 552B backbone plus a 196B Engram conditional-memory module make a 763B checkpoint of 510 GB (FP8 with FP4 routed experts, read on Hugging Face on 2026-10-27); 8B parameters are active at prefill and 16B at decode; 1M-token context. The update V4.1-Flash-1017 of 2026-10-17 keeps the architecture and price (77.0 on DeepSWE v1.1, per DeepSeek). DeepSeek has announced a V4.1 Pro but not released it; V4 Pro remains served.

GPUBank global report, 2026-10-25, §3.4 (internal, sourced)
https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017
https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash
https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017/raw/main/config.json
https://huggingface.co/api/models/deepseek-ai/DeepSeek-V4.1-Flash-1017?blobs=true
https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash/raw/main/README.md
https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash/raw/main/config.json

Memory · 1 precision · 6 context lengths

How 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	1M context
INT4-packed + FP8/BF16published checkpoint	510 GB	613 GB	613 GB	616 GB	624 GB	634 GB	699 GB
KV cacheone sequence, FP16 cache, included above	—	—	1 GB	3 GB	11 GB	22 GB	86 GB

Cross-checked on Hugging Face Oct 28, 2026: 763.2B parameters in the safetensors headers, 510.3 GB of weight files, attention layout from config.json (40 layers, 1 KV heads × 512). 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 = 2048 bytes per token per layer × 40 full-attention layers (FP16 cache, one sequence).config.json ↗

Smallest configuration per GPU · 110 GPUs · cheapest eligible on-demand offer per GPU count, every provider on file

What 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
RTX PRO 6000Best fit	8× 96 GB768 GB total	80%	$7.78	$5,676	UpCloudTrust 75 · read Nov 1 · receipt · also Vast.ai $10.14, Lium $12.80
Gaudi 2	8× 96 GB768 GB total	80%	$8.27	$6,036	Cyfuture AInot scored · read Nov 1 · receipt
A100 80GB	8× 80 GB640 GB total	96% tight	$9.82	$7,166	Edgevananot scored · read Nov 1 · receipt · also GPU.ai $10.00, io.net $10.50
MI300X	4× 192 GB768 GB total	80%	$11.31	$8,257	Cyfuture AInot scored · read Nov 1 · receipt · also Hot Aisle $11.96, AtmosCompute $13.38
MI325X	4× 256 GB1024 GB total	60%	$12.29	$8,972	Cyfuture AInot scored · read Nov 1 · receipt
H100	8× 80 GB640 GB total	96% tight	$13.67	$9,982	GcoreTrust 80 · read Nov 1 · receipt · also Lium $14.36, Airon AI $15.04
B200	4× 180 GB720 GB total	85%	$16.46	$12,013	UpCloudTrust 75 · read Nov 1 · receipt · also Deep Infra $19.56, Lambda $27.16
H200	8× 141 GB1128 GB total	54%	$21.04	$15,359	Airon AInot scored · read Nov 1 · receipt · also Geodd $21.76, Vast.ai $22.11
B300	4× 288 GB1152 GB total	53%	$23.29	$17,004	UpCloudTrust 75 · read Nov 1 · receipt · also FPT AI Factory $27.96, Vast.ai $37.50
MI355X	8× 288 GB2304 GB total	27%	$23.60	$17,228	TensorWaveTrust 64 · read Nov 1 · receipt
H200 NVL	8× 141 GB1128 GB total	54%	$26.09	$19,042	IONOSnot scored · read Nov 1 · receipt · also Geodd $27.12, Vast.ai $39.47
B200 CC	4× 180 GB720 GB total	85%	$36.19	$26,419	VerdaTrust 81 · read Nov 1 · receipt
B300 CC	4× 288 GB1152 GB total	53%	$46.31	$33,806	VerdaTrust 81 · read Nov 1 · receipt
A800 80GB PCIe	8× 80 GB640 GB total	96% tight	$50.80	$37,088	Volcano Enginenot scored · read Nov 1 · receipt
H100 NVL	8× 94 GB752 GB total	81%	$51.79	$37,809	Open Telekom Cloudnot scored · read Nov 1 · receipt · also Cloud.ru $52.63
GB300	4× 288 GB1152 GB total	53%	$52.21	$38,113	VerdaTrust 81 · read Nov 1 · receipt

Throughput estimates · internal study · 2026-10-04

Tokens 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 the study's own estimate at the prices it read on 2026-10-04; it is not recomputed here.

GPU	Cards	Precision	tok/s per user	per 1M output tokens · study estimate
MI300X	4	FP4	63.6	$1.55
B200	4	FP4	108	$1.91
H200	4	FP4	81.8	$1.95
A100 80GB	8	FP4	55.3	$2.12
H100 SXM	8	FP4	67.3	$2.59
RTX PRO 6000	8	FP4	43.5	$3.67
H100 PCIe	8	FP4	40.7	$4.23

cards, precision and speed estimated from memory bandwidth and model size (internal study); cost recomputed by the site from verified prices · cheapest per output at study prices: MI300XWhy estimates, and their limits →

DeepSeek V4.1 Flash · related

Same publisher or category

What is documented about DeepSeek V4.1 Flash

How much VRAM it needs

What it costs to run, GPU by GPU

Tokens per second, and what a million costs

Models close to this one