DeepSeekDeepSeek V4 Pro
DeepSeek V4 Pro is a DeepSeek open-weights model, released 13 Aug 2026. Its best configuration (max reasoning effort) ranks #72 of 756 on the BenchLeader Index at 62.9 ±3.3, in the upper half. The ± is the point: 147 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in instruction following (74) and lowest in maths (56). At $0.990 per million tokens blended it is mid-priced. Output speed of 107 tokens per second puts it faster than most, with a first answer in 42.6 s. It has been measured at 6 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 29 Sept 2026.
- Blended price
- $0.990/M
- $0.660 in · $1.98 out
- Output speed
- 107 tok/s
- measured by Artificial Analysis
- First answer
- 43 s
- first token 2.35 s
- Context
- 1M
- Full answer
- 47 s
- median, reasoning included
- Cost per run
- $0.674
- one full Intelligence Index run
- Released
- 13 Aug 2026
- Overall index63
- Reasoning65
- Coding58
- Agents & tools64
- Maths56
- Knowledge60
- Instruction following74
- Long context66
- Composite71
Reasoning-effort configurations
The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| no reasoning | 49.4 | #382 | 153 tok/s | 2.03 s | $0.0009 | Agents & tools 65 · Coding 35 · Composite 53 · Instruction following 48 · Knowledge 56 · Long context 51 · Maths 44 · Reasoning 41 |
| low | – | – | 35 tok/s | 2.35 s | $0.0009 | Reasoning 60 |
| high | 56.6 | #185 | 110 tok/s | 20 s | $0.0009 | Agents & tools 57 · Coding 41 · Composite 64 · Instruction following 70 · Knowledge 57 · Long context 60 · Reasoning 59 |
| xhigh | – | – | 35 tok/s | 2.35 s | $0.0009 | |
| maxbest | 62.9 | #72 | 107 tok/s | 43 s | $0.0009 | Agents & tools 64 · Coding 58 · Composite 71 · Instruction following 74 · Knowledge 60 · Long context 66 · Maths 56 · Reasoning 65 |
| not stated | 53.9 | #255 | 35 tok/s | 2.35 s | $0.0009 | Agents & tools 43 · Coding 54 · Composite 57 · Human preference 58 · Maths 60 · Reasoning 57 |
How its index has moved
8 Sept 2026 to 5 Oct 2026- max
- high
- not stated
- no reasoning
The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.
Benchmark results
One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Metadata
| Benchmark | no reasoning | low | high | xhigh | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|
| AA Openness Indexnot in index | – | – | – | – | 44.4#26 | – | Artificial Analysis |
Reasoning
| Benchmark | no reasoning | low | high | xhigh | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|
| GPQA Diamond | 73.2%#152 | – | – | – | 91.7%#28 | – | Epoch AI Benchmarking Hub | |
| LMArena Hard Prompts | – | – | – | – | – | 1480#65 | LMArena | |
| LiveBench Reasoningnot in index | – | – | – | – | – | 85.8%#33 | LiveBench | |
| GPQA Diamond (AA)not in index | 71.7%#278 | – | 90.5%#59 | – | 92.8%#28 | – | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 10.6%#293 | – | 35.2%#106 | – | 41.0%#72 | – | Artificial Analysis | |
| GPQA Diamond (Vals)not in index | – | – | – | – | 92.4%#16 | – | Vals AI | |
| Kagi LLM Benchmark | – | – | – | – | – | 53.5%#70 | Kagi LLM Benchmark | |
| ARC-AGI-1 | 13.0%#196 | 90.5%#65 | 87.2%#79 | – | 90.0%#69 | – | ARC Prize | |
| ARC-AGI-2 | 0.8%#196 | 56.3%#83 | 59.7%#77 | – | 61.3%#72 | – | ARC Prize | |
| CritPt | 0.9%#217 | – | 10.0%#105 | – | 18.0%#63 | – | Artificial Analysis | |
| Chess Puzzlesnot in index | 7.0%#143 | – | – | – | 47.0%#14 | – | Epoch AI Benchmarking Hub | |
| Mystery Game Puzzlesnot in index | 17.0%#70 | – | – | – | 43.0%#13 | – | Epoch AI Benchmarking Hub | |
| LMCAnot in index | – | – | – | – | 41.2%#86 | 45.5%#71 | Epoch AI Benchmarking Hub | |
| DTBenchnot in index | – | – | – | – | 90.7%#64 | 93.9%#47 | Epoch AI Benchmarking Hub | |
| ForecastBenchnot in index | – | – | – | – | – | 56.1%#68 | ForecastBench |
Coding
Agents & tools
Maths
| Benchmark | no reasoning | low | high | xhigh | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|
| FrontierMath Tiers 1–3 | – | – | – | – | 64.6%#31 | – | Epoch AI Benchmarking Hub | |
| FrontierMath Tier 4 | – | – | – | – | 26.8%#38 | – | Epoch AI Benchmarking Hub | |
| OTIS Mock AIME | 46.7%#178 | – | – | – | 98.6%#19 | – | Epoch AI Benchmarking Hub | |
| ProofBench | – | – | – | – | 16.0%#54 | 50.0%#28 | Vals AI | |
| LiveBench Mathematicsnot in index | – | – | – | – | – | 95.1%#16 | LiveBench | |
| AIME 2026 | – | – | – | – | 96.7%#6 | – | MathArena | |
| HMMT February 2026 | – | – | – | – | 93.9%#10 | – | MathArena | |
| MathArena Apex | – | – | – | – | 28.1%#10 | – | MathArena | |
| LMArena Maths | – | – | – | – | – | 1445#78 | LMArena |
Knowledge
| Benchmark | no reasoning | low | high | xhigh | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|
| SimpleQA Verified | – | – | – | – | 52.9%#21 | – | Epoch AI Benchmarking Hub | |
| LiveBench Data Analysisnot in index | – | – | – | – | – | 79.2%#17 | LiveBench | |
| AA-Omniscience | -12.4#194 | – | -10.6#181 | – | 0.8#116 | – | Artificial Analysis | |
| MMLU-Pro | – | – | – | – | 87.3%#32 | – | Vals AI | |
| LegalBench | – | – | – | – | 82.4%#69 | – | Vals AI | |
| CorpFin | – | – | – | – | 65.4%#28 | – | Vals AI | |
| TaxEval | – | – | – | – | 73.1%#52 | – | Vals AI | |
| CaseLaw v2not in index | – | – | – | – | 59.4%#27 | – | Vals AI | |
| Excel Modeling Benchmarknot in index | – | – | – | – | 52.8%#44 | – | Vals AI | |
| MedCodenot in index | – | – | – | – | 42.5%#47 | – | Vals AI | |
| MedScribenot in index | – | – | – | – | 80.2%#54 | – | Vals AI | |
| Public Benefits Benchnot in index | – | – | – | – | 62.9%#21 | – | Vals AI | |
| DrugDiscoveryBenchnot in index | – | – | – | 31.7%#13 | – | – | Scale AI SEAL | |
| AA-Omniscience: accuracynot in index | 41.5%#108 | – | 41.4%#109 | – | 49.1%#65 | – | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in index | 12.2%#360 | – | 11.3%#372 | – | 5.9%#489 | – | Artificial Analysis |
Instruction following
| Benchmark | no reasoning | low | high | xhigh | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|
| LiveBench Languagenot in index | – | – | – | – | – | 82.1%#25 | LiveBench | |
| LiveBench Instruction Followingnot in index | – | – | – | – | – | 67.7%#34 | LiveBench | |
| IFBench | 45.8%#198 | – | 71.3%#57 | – | 76.5%#20 | – | Artificial Analysis | |
| LMArena Instruction Followingnot in index | – | – | – | – | – | 1452#62 | LMArena |
Human preference
| Benchmark | no reasoning | low | high | xhigh | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|
| LMArena Text | – | – | – | – | – | 1458#62 | LMArena | |
| EQ-Bench 4 | – | – | – | – | – | 1166#17 | EQ-Bench | |
| LMArena Creative Writingnot in index | – | – | – | – | – | 1446#48 | LMArena | |
| LMArena Multi-turnnot in index | – | – | – | – | – | 1475#47 | LMArena | |
| LMArena Longer Queriesnot in index | – | – | – | – | – | 1472#58 | LMArena |
Long context
| Benchmark | no reasoning | low | high | xhigh | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|
| AA-LCR | 53.0%#288 | – | 70.3%#195 | – | 80.3%#63 | – | Artificial Analysis | |
| MLCRnot in index | – | – | – | – | 17.8%#22 | – | Artificial Analysis |
Composite
| Benchmark | no reasoning | low | high | xhigh | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|
| Epoch Capabilities Indexnot in index | – | – | – | – | – | 155.4#26 | Epoch AI Benchmarking Hub | |
| LiveBench | – | – | – | – | – | 77.4%#22 | LiveBench | |
| AA Intelligence Index | 20.8#202 | – | 30.1#107 | – | 36#75 | – | Artificial Analysis | |
| Vals Indexnot in index | – | – | – | – | 47.6#30 | – | Vals AI |
Where to run it
Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| Relace | 79 tok/s | 0.92 s | $0.207 | $4.20 | 1.0M | fp4 |
| Azure | 57 tok/s | 1.63 s | $1.91 | $3.83 | 1.0M | – |
| NovitaAI | 54 tok/s | 2.48 s | $1.60 | $3.20 | 1.0M | fp8 |
| Baidu Qianfan | 52 tok/s | 1.05 s | $1.69 | $3.38 | 1.0M | fp8 |
| Parasail | 45 tok/s | 0.70 s | $0.450 | $3.48 | 1.0M | fp8 |
| AtlasCloud | 43 tok/s | 1.53 s | $1.68 | $3.38 | 1.0M | fp4 |
| NextBit | 43 tok/s | 4.21 s | $1.74 | $3.48 | 1.0M | fp8 |
| Alibaba Cloud Int. | 43 tok/s | 1.82 s | $1.42 | $2.83 | 1M | fp8 |
| Cloudflare | 42 tok/s | 2.64 s | $1.15 | $2.55 | 1.0M | – |
| DeepInfra | 39 tok/s | 1.46 s | $1.30 | $2.60 | 1.0M | fp8 |
| SiliconFlow | 39 tok/s | 2.27 s | $1.50 | $3.13 | 1.0M | fp8 |
| Venice | 39 tok/s | 1.53 s | $1.65 | $3.30 | 1M | – |
| StreamLake | 39 tok/s | 2.87 s | $0.209 | $0.418 | 1.0M | fp8 |
| Reka AI | 26 tok/s | 0.82 s | $1.20 | $6.00 | 1.0M | – |
| GMICloud | 20 tok/s | 2.95 s | $0.957 | $1.91 | 1.0M | fp8 |
| DigitalOcean | 12 tok/s | 2.47 s | $1.04 | $2.09 | 1.0M | – |
Price history
Listed price per 1M tokens over time, as recorded by OpenRouter for the provider with the longest history.
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache at $0.022 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0009 | $0.0007 | 45.5 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0091 | $0.0034 | 48.3 s |
| Code edit | 6,000 / 1,500 | $0.0069 | $0.0041 | 56.7 s |
| Agentic coding session | 60,000 / 4,000 | $0.048 | $0.019 | 1.3 min |
| Structured extraction | 2,000 / 200 | $0.0017 | $0.0008 | 44.5 s |
See also
Data as of 5 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “DeepSeek V4 Pro: benchmarks, pricing, speed and rank”, https://www.benchleader.com/models/deepseek-v4-pro, data as of 5 Oct 2026.