DeepSeek V4 Pro (0813)
- Input
- $1.32/M tok
- Output
- $3.96/M tok
- Blended
- $1.98/M tok · 3:1
- Speed
- 97tok/s
- Context
- 1Mtokens
- Cost per task
- ~$1.7/taskest.
Open weights. Peak-hour price; off-peak is 50% off.
Rankings
Coding
Verified#53 / 137rank by quality56.8±4.5quality / 1004/7 benchmarks measuredWriting & chat
Verified#27 / 162rank by quality70.6±6.0quality / 1002/5 benchmarks measuredHard reasoning
Verified#39 / 218rank by quality70.9±3.7quality / 1003/4 benchmarks measuredAgents
Verified#30 / 187rank by quality69.7±5.9quality / 1003/7 benchmarks measured
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
- 24× H200 141GB
- FP8 weights
- Throughput
- ~29,781 tok/s
- many requests batched
- Price
- $0.39–3.23 /M tok
- busy → light use
- On your own machine
- Too large
- needs data-center GPUs
Assumptions (5)
- FP8 weights (1600B params, 49B active per token) + 40% KV-cache headroom ≈ 2240 GB
- 24× H200 141GB at $3.59–$7.91/GPU-hour on-demand (2026-09-24)
- ~29,781 output tok/s aggregate at batch 1260 (bandwidth-bound); MoE compute scales with active params
- Blended 3:1 input:output; prefill ~194,649 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
~#4 of 21 (estimated)~81.3%
- #1Claude Opus 596.0%
- #2Claude Sonnet 585.2%
- #3Beam 1.0 501B A23B81.6%
- ~#4DeepSeek V4 Pro (0813)Estimated81.3%
- #4Gemini 3.1 Pro80.6%
No published score — estimated from related benchmarks
SWE-bench Pro (Public)Estimated
~#11 of 24 (estimated)~62.6%
- #1Claude Fable 5.293.0%
- #2Claude Opus 5.589.9%
- #10Claude Sonnet 563.2%
- ~#11DeepSeek V4 Pro (0813)Estimated62.6%
- #11Nex-N2.5-Pro61.2%
No published score — estimated from related benchmarks
DeepSWE v1.1Verified
#34 of 4557.2%
- #1Gemini 4 Argon78.8%
- #2DeepSeek V4.1 Flash 101777.0%
- #33Qwen3.8-Omni-Flash57.8%
- #34DeepSeek V4 Pro (0813)57.2%
- #35Qwen3.8 2.4T A95B56.6%
artificialanalysis.ai · 2026-09-24
Terminal-Bench 4.0Verified
#34 of 10114.1%
- #1Kimi K3.1 Pro67.2%
- #2Claude Sonnet 5.563.6%
- #34Claude Sonnet 514.1%
- #34DeepSeek V4 Pro (0813)14.1%
- #36GPT-5.5 Instant (June 2026)12.6%
artificialanalysis.ai · 2026-09-24
LiveCodeBenchEstimated
~#11 of 28 (estimated)~80.6%
- #1Qwen3.8-Omni-Flash92.6%
- #2Qwen3.8 Flash-Next91.9%
- #10Ternary Bonsai 27B82.8%
- ~#11DeepSeek V4 Pro (0813)Estimated80.6%
- #11Gemma 4 31B80.0%
No published score — estimated from related benchmarks
Writing & chat
IFBenchEstimated
~#31 of 140 (estimated)~72.2%
- #1MiniMax-M382.9%
- #2Qwen3.8 2.4T A95B82.8%
- #30Gemma 4 26B A4B72.4%
- ~#31DeepSeek V4 Pro (0813)Estimated72.2%
- #31Nemotron 3 Super 120B A12B71.5%
No published score — estimated from related benchmarks
LMArena Text, Non-English (Elo)Verified
#18 of 561448 Elo
- #1Gemini 4 ArgonFew votes1512
- #2Claude Opus 5.51497
- #17GPT-5.6 Terra1456
- #18DeepSeek V4 Pro (0813)1448
- #19Claude Sonnet 51447
lmarena.ai · 2026-09-24
LMArena Text (Elo)Verified
#18 of 561463 Elo
- #1Gemini 4 Argon1525
- #2Claude Opus 5.51507
- #17GPT-5.6 Terra1466
- #18DeepSeek V4 Pro (0813)1463
- #19Claude Sonnet 51461
lmarena.ai · 2026-09-24
EQ-Bench Creative Writing v3 (Elo)Estimated
~#9 of 26 (estimated)~1913 Elo
- #1GPT-6 Astra2173
- #2Claude Fable 5.12162
- #8Grok 4.72007
- ~#9DeepSeek V4 Pro (0813)Estimated1913
- #9Muse Spark 1.31906
No published score — estimated from related benchmarks
Hard reasoning
GPQA DiamondVerified
#13 of 19292.8%
- #1GPT-6 Astra96.1%
- #2Gemini 3.8 Flash95.3%
- #12MiniMax-M392.9%
- #13DeepSeek V4 Pro (0813)92.8%
- #13Qwen3.8 Max (0902)92.8%
artificialanalysis.ai · 2026-09-24
Humanity's Last ExamVerified
#33 of 19841.0%
- #1Claude Fable 5.274.3%
- #2Claude Opus 5.561.4%
- #32Claude Sonnet 541.3%
- #33DeepSeek V4 Pro (0813)41.0%
- #34GLM-5.3 Flash39.9%
artificialanalysis.ai · 2026-09-24
AIME (latest)Estimated
~#4 of 34 (estimated)~96.4%
- #1GLM-5.3 NVFP499.5%
- #2Beam 1.0 501B A23B97.7%
- #3Inkling97.1%
- ~#4DeepSeek V4 Pro (0813)Estimated96.4%
- #4Ternary Bonsai 2 27B95.8%
No published score — estimated from related benchmarks
Agents
τ²-bench (Telecom)Estimated
~#6 of 120 (estimated)~94.8%
- #1Step 3.7 Flash98.5%
- #2DeepSeek-V4-Pro-0813-nvfp4-DSpark98.3%
- #5Qwen3.6 35B A3B95.3%
- ~#6DeepSeek V4 Pro (0813)Estimated94.8%
- #6Mistral Medium 3.594.2%
No published score — estimated from related benchmarks
OSWorld-VerifiedEstimated
~#8 of 15 (estimated)~79.9%
- #1Kimi K384.8%
- #2Qwen3.8 27B84.3%
- #7Muse Spark Open80.4%
- ~#8DeepSeek V4 Pro (0813)Estimated79.9%
- #8GPT-5.578.7%
No published score — estimated from related benchmarks
OSWorld 2.0Estimated
~#3 of 5 (estimated)~50.9%
No published score — estimated from related benchmarks
BrowseCompEstimated
~#8 of 22 (estimated)~86.4%
- #1Atria Dawn Preview92.5%
- #2GPT-6 Astra91.5%
- #7GPT-5.6 Terra87.5%
- ~#8DeepSeek V4 Pro (0813)Estimated86.4%
- #8Gemini 3.1 Pro85.9%
No published score — estimated from related benchmarks
GDPval-AA (Elo)Verified
#23 of 911441 Elo
- #1Claude Opus 5.51866
- #2Muse Wonder Full1857
- #22Claude Sonnet 51449
- #23DeepSeek V4 Pro (0813)1441
- #24Gemini 3.8 Flash1436
artificialanalysis.ai · 2026-09-24