DeepSeek V4.1 Flash
- Input
- $0.30/M tok
- Output
- $1.20/M tok
- Blended
- $0.525/M tok · 3:1
- Speed
- 232tok/s
- Context
- 1Mtokens
- Cost per task
- ~$0.51/taskest.
Open weights (MIT). Peak-hour price; off-peak is 50% off.
Rankings
Coding
Mixed#29 / 137rank by quality67.7±4.7quality / 1004/7 benchmarks measuredSpeed frontierWriting & chat
Verified#40 / 162rank by quality65.2±4.3quality / 1003/5 benchmarks measuredSpeed frontierVision
Verified#50 / 98rank by quality55.2±4.8quality / 1003/6 benchmarks measuredSpeed frontierHard reasoning
Mixed#38 / 218rank by quality71.4±4.0quality / 1003/4 benchmarks measuredSpeed frontierAgents
Verified#19 / 187rank by quality78.5±7.8quality / 1002/7 benchmarks measuredSpeed frontier
Self-hostingEstimated
How we estimate →What it would cost to run these open weights yourself on rented GPUs, compared with the API price above.
- Hardware
- 8× B200 180GB
- FP8 weights
- Throughput
- ~39,657 tok/s
- many requests batched
- Price
- $0.14–1.22 /M tok
- busy → light use
- On your own machine
- Mac Studio M5 Ultra 512GB (int4, ~110 tok/s single-stream)
- single consumer GPU or Mac
Assumptions (5)
- FP8 weights (552B params, 16B active per token) + 40% KV-cache headroom ≈ 773 GB
- 8× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
- ~39,657 output tok/s aggregate at batch 1135 (bandwidth-bound); MoE compute scales with active params
- Blended 3:1 input:output; prefill ~531,563 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
SWE-bench VerifiedEstimated
~#3 of 21 (estimated)~83.6%
- #1Claude Opus 596.0%
- #2Claude Sonnet 585.2%
- ~#3DeepSeek V4.1 FlashEstimated83.6%
- #3Beam 1.0 501B A23B81.6%
No published score — estimated from related benchmarks
SWE-bench Pro (Public)Estimated
~#7 of 24 (estimated)~66.5%
- #1Claude Fable 5.293.0%
- #2Claude Opus 5.589.9%
- #6Intern-S2-397B68.5%
- ~#7DeepSeek V4.1 FlashEstimated66.5%
- #7Beam 1.0 501B A23B66.1%
No published score — estimated from related benchmarks
DeepSWE v1.1Lab-reported
#6 of 4574.2%
- #1Gemini 4 Argon78.8%
- #2DeepSeek V4.1 Flash 101777.0%
- #5Claude Fable 5.274.3%
- #6DeepSeek V4.1 Flash74.2%
- #7MiniMax-M3.1-Flash73.7%
huggingface.co · 2026-09-10
Terminal-Bench 4.0Verified
#24 of 10126.8%
- #1Kimi K3.1 Pro67.2%
- #2Claude Sonnet 5.563.6%
- #22GLM-5.3 Flash32.8%
- #24DeepSeek V4.1 Flash26.8%
- #24Mistral Large 426.8%
artificialanalysis.ai · 2026-09-24
LiveCodeBenchEstimated
~#11 of 28 (estimated)~80.8%
- #1Qwen3.8-Omni-Flash92.6%
- #2Qwen3.8 Flash-Next91.9%
- #10Ternary Bonsai 27B82.8%
- ~#11DeepSeek V4.1 FlashEstimated80.8%
- #11Gemma 4 31B80.0%
No published score — estimated from related benchmarks
Writing & chat
IFBenchEstimated
~#34 of 140 (estimated)~71.3%
- #1MiniMax-M382.9%
- #2Qwen3.8 2.4T A95B82.8%
- #33o371.4%
- ~#34DeepSeek V4.1 FlashEstimated71.3%
- #34GPT-5.6 Terra71.2%
No published score — estimated from related benchmarks
LMArena Text, Non-English (Elo)Verified
#16 of 561459 Elo
- #1Gemini 4 ArgonFew votes1512
- #2Claude Opus 5.51497
- #15MiMo-V2.6-Pro1461
- #16DeepSeek V4.1 Flash1459
- #17GPT-5.6 Terra1456
lmarena.ai · 2026-10-02
LMArena Text (Elo)Verified
#16 of 561474 Elo
- #1Gemini 4 Argon1525
- #2Claude Opus 5.51507
- #15GLM-5.3 Flash1475
- #16DeepSeek V4.1 Flash1474
- #17GPT-5.6 Terra1466
lmarena.ai · 2026-10-02
Vision
MMMU-ProVerified
#28 of 8777.0%
- #1Claude Opus 5.587.7%
- #2GPT-6 Astra86.9%
- #26Qwen3.5 397B A17B77.3%
- #28DeepSeek V4.1 Flash77.0%
- #29Intern-S2-Preview (35B-A3B)76.9%
artificialanalysis.ai · 2026-09-24
CharXiv ReasoningEstimated
~#8 of 12 (estimated)~79.4%
- #1Gemini 3.8 Flash86.2%
- #2Kimi K384.8%
- #7Nemotron 3.5 Super VL NVFP479.9%
- ~#8DeepSeek V4.1 FlashEstimated79.4%
- #8Muse Glimmer78.8%
No published score — estimated from related benchmarks
OmniDocBench v1.5Estimated
~#5 of 8 (estimated)~84.9%
- #1MiniMax-M391.6%
- #2Kimi K391.1%
- #4Gemini 3.1 Pro85.3%
- ~#5DeepSeek V4.1 FlashEstimated84.9%
- #5Claude Haiku 4.579.6%
No published score — estimated from related benchmarks
GDP.pdf (AA)Verified
#29 of 4712.8%
- #1Muse Wonder Full32.6%
- #2GPT-6 Astra31.0%
- #28Claude Sonnet 513.2%
- #29DeepSeek V4.1 Flash12.8%
- #29Inkling12.8%
artificialanalysis.ai · 2026-09-24
ChartographyVerified
#21 of 2411.6%
- #1Claude Fable 5.273.4%
- #2GPT-6 Astra71.0%
- #20Grok 4.714.7%
- #21DeepSeek V4.1 Flash11.6%
- #22Gemini 3.5 Flash-Lite10.2%
surgehq.ai · 2026-09-24
Hard reasoning
GPQA DiamondLab-reported
#28 of 19290.9%
- #1GPT-6 Astra96.1%
- #2Gemini 3.8 Flash95.3%
- #27Qwen3.8-Omni-Flash91.0%
- #28DeepSeek V4.1 Flash90.9%
- #29Beam 1.0 501B A23B90.6%
huggingface.co · 2026-09-10
Humanity's Last ExamVerified
#36 of 19839.2%
- #1Claude Fable 5.274.3%
- #2Claude Opus 5.561.4%
- #35Ling 3.1 Flash39.4%
- #36DeepSeek V4.1 Flash39.2%
- #37MiniMax-M339.0%
artificialanalysis.ai · 2026-09-24
AIME (latest)Estimated
~#5 of 34 (estimated)~95.6%
- #1GLM-5.3 NVFP499.5%
- #2Beam 1.0 501B A23B97.7%
- #4Ternary Bonsai 2 27B95.8%
- ~#5DeepSeek V4.1 FlashEstimated95.6%
- #5Ripple-65B Preview95.0%
No published score — estimated from related benchmarks
Agents
τ²-bench (Telecom)Estimated
~#12 of 120 (estimated)~92.1%
- #1Step 3.7 Flash98.5%
- #2DeepSeek-V4-Pro-0813-nvfp4-DSpark98.3%
- #11Ring-2.6-1T92.4%
- ~#12DeepSeek V4.1 FlashEstimated92.1%
- #12MiMo-V2.590.6%
No published score — estimated from related benchmarks
OSWorld-VerifiedEstimated
~#7 of 15 (estimated)~80.7%
- #1Kimi K384.8%
- #2Qwen3.8 27B84.3%
- #6MiMo-V2.6-Flash80.8%
- ~#7DeepSeek V4.1 FlashEstimated80.7%
- #7Muse Spark Open80.4%
No published score — estimated from related benchmarks
OSWorld 2.0Estimated
~#3 of 5 (estimated)~57.5%
No published score — estimated from related benchmarks
BrowseCompEstimated
~#8 of 22 (estimated)~86.5%
- #1Atria Dawn Preview92.5%
- #2GPT-6 Astra91.5%
- #7GPT-5.6 Terra87.5%
- ~#8DeepSeek V4.1 FlashEstimated86.5%
- #8Gemini 3.1 Pro85.9%
No published score — estimated from related benchmarks
GDPval-AA (Elo)Verified
#15 of 911600 Elo
- #1Claude Opus 5.51866
- #2Muse Wonder Full1857
- #14Qwen3.8 Flash-Next1612
- #15DeepSeek V4.1 Flash1600
- #16Qwen3.8 2.4T A95B1597
artificialanalysis.ai · 2026-09-24
AutomationBench-AAVerified
#4 of 8468.9%
- #1Claude Fable 5.276.4%
- #2Muse Wonder Full71.3%
- #3Claude Opus 5.569.5%
- #4DeepSeek V4.1 Flash68.9%
- #5GPT-6 Astra68.5%
artificialanalysis.ai · 2026-09-24
τ-Bench Banking (AA)Estimated
~#15 of 91 (estimated)~40.1%
- #1Muse Spark 1.350.5%
- #2GLM-5.350.3%
- #14GPT-5.6 Terra40.2%
- ~#15DeepSeek V4.1 FlashEstimated40.1%
- #15DeepSeek V4 Pro (0813)39.6%
No published score — estimated from related benchmarks