Skip to content
Model Pareto

DeepSeek V4.1 Flash NVFP4

NVIDIAOpen weightsNVFP4quantized from DeepSeek V4.1 Flashhuggingface.co ↗
Input
—
Output
—
Speed
~160tok/sest.
Context
1Mtokens

Rankings

Self-hostingEstimated

How we estimate →

What it would cost to run these open weights yourself on rented GPUs. No API sells this model, so this is its price on the chart.

Hardware
4× B200 180GB
NVFP4 weights
Throughput
~17,975 tok/s
many requests batched
Price
$0.15–1.32 /M tok
busy → light use
On your own machine
Mac Studio M5 Ultra 512GB (NVFP4, ~93 tok/s single-stream)
single consumer GPU or Mac
Assumptions (5)
  • NVFP4 weights (552B params, 16B active per token) + 40% KV-cache headroom ≈ 435 GB
  • 4× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
  • ~17,975 output tok/s aggregate at batch 514 (bandwidth-bound); MoE compute scales with active params
  • Blended 3:1 input:output; prefill ~265,781 tok/s
  • Low = 75% utilization at the low GPU price; high = 20% at the high price

Benchmark scores

~ italic, dashed = no published score yet, estimated from related benchmarks.

Coding

~#2 of 21 (estimated)~86.1%
  1. #1Claude Opus 596.0%
  2. ~#2DeepSeek V4.1 Flash NVFP4Estimated86.1%
  3. #2Claude Sonnet 585.2%
No published score — estimated from related benchmarks
DeepSWE v1.1Estimated
~#24 of 46 (estimated)~63.9%
  1. #1Gemini 4 Argon78.8%
  2. #2DeepSeek V4.1 Flash 101777.0%
  3. #22Hy4 preview64.3%
  4. ~#24DeepSeek V4.1 Flash NVFP4Estimated63.9%
  5. #24GPT-6 Luna63.7%
No published score — estimated from related benchmarks
LiveCodeBenchEstimated
~#3 of 28 (estimated)~90.7%
  1. #1Qwen3.8-Omni-Flash92.6%
  2. #2Qwen3.8 Flash-Next91.9%
  3. ~#3DeepSeek V4.1 Flash NVFP4Estimated90.7%
  4. #3Qwen3.8 27B90.3%
No published score — estimated from related benchmarks
~#11 of 26 (estimated)~49.6
  1. #1Claude Opus 5.561.7
  2. #2Claude Opus 559.7
  3. #10Muse Wonder52.0
  4. ~#11DeepSeek V4.1 Flash NVFP4Estimated49.6
  5. #11GLM-5.348.9
No published score — estimated from related benchmarks

Writing & chat

Vision

MMMUEstimated
~#2 of 3 (estimated)~71.3%
  1. #1Claude Haiku 4.573.2%
  2. ~#2DeepSeek V4.1 Flash NVFP4Estimated71.3%
  3. #2Llama 4 Maverick69.4%
No published score — estimated from related benchmarks
~#4 of 8 (estimated)~86.8%
  1. #1MiniMax-M391.6%
  2. #2Kimi K391.1%
  3. #2Qwen3.8 27B91.1%
  4. ~#4DeepSeek V4.1 Flash NVFP4Estimated86.8%
  5. #4Gemini 3.1 Pro85.3%
No published score — estimated from related benchmarks
GDP.pdf (AA)Estimated
~#20 of 48 (estimated)~18.3%
  1. #1Muse Wonder Full32.6%
  2. #2GPT-6 Astra31.0%
  3. #19MiMo-V2.6-Pro19.2%
  4. ~#20DeepSeek V4.1 Flash NVFP4Estimated18.3%
  5. #20Gemini 3.1 Pro17.8%
No published score — estimated from related benchmarks
ChartographyEstimated
~#10 of 25 (estimated)~29.5%
  1. #1Claude Fable 5.273.4%
  2. #2GPT-6 Astra71.0%
  3. #9GPT-5.531.0%
  4. ~#10DeepSeek V4.1 Flash NVFP4Estimated29.5%
  5. #10GPT-6 Luna29.1%
No published score — estimated from related benchmarks

Hard reasoning