Models · DeepSeek · Out sinceReleased 13 Aug 2026
DeepSeek V4 Pro (0813)#26 for coding.#26 for coding, best at max effort.
Available on OpenRouterOn OpenRouterAvailable nowGenerally availableCan run on your own serversOpen weights
DeepSeek V4 Pro (0813) is made by DeepSeek. Among the models we track it ranks #26 for coding, #29 for writing, #33 for research and analysis. It's mid-priced to use. Its makers have published it, so you can run it on your own servers.
GA release of V4-Pro (1.6T total / 49B active, MIT), served as API model 'deepseek-v4-pro'. List price = peak rate; off-peak 50% ($0.66/$1.98). Vendor guidance: low for simple tasks, high for daily agent workflows, max for complex tasks. Native OpenAI Responses API (Codex).
- 35.8 / 100 · #29
- 44.4 / 100 · #33
- 36.3 / 100 · #26
- Mid-priced$1.32 / $3.96
- $1.98
- 1M
- 393K
- 55 (40 independent40 indep.)
- 13 Aug 2026
- DeepSeek
- text
Scores are out of 100 for each area, compared with every model we track; “#” is its rank. Context is how much text it can read at once (1M is roughly 700,000 words). Blended price mixes the cost of what you send and what it writes back.
No reasoningLowHighMax
Highlighted: where it did best for coding (Maximum thinking).
Highlighted: dominant setting in its coding composite.
OpenRouter is a service that gives access to many AI models in one place. This model is listed there as deepseek/deepseek-v4-pro-0813.
Route it as deepseek/deepseek-v4-pro-0813 at $1.32 in / $3.96 out per 1M tokens, 1M context. Listed since 12 Aug 2026.
frequency_penaltyinclude_reasoninglogit_biaslogprobsmax_tokensmin_ppresence_penaltyreasoningreasoning_effortrepetition_penaltyresponse_formatseedstopstructured_outputstemperaturetool_choicetoolstop_ktop_logprobstop_p
Run it yourself
Self-hosting
On your own servers,Licence, sizeunder your own control.and hardware.
- Yes
- 1600B / 49B
Needs several AI servers.
> 1.1 TB at 8-bit (multi-node). Weights ≈ 1680 GB at 8-bit, 880 GB at 4-bit (+10–30% for KV cache). MoE: 49B active per token.
Languages: Languages: Not stated.
Training it further: Fine-tuning: No recipe. DeepSeek-V4-Pro-Base (MIT, FP8, ~1.6T) is published - the only frontier-scale open base model, but impractical to fine-tune in-house.
Quantisations: fp4+fp8 mixed (native: FP4 MoE experts, FP8 elsewhere), nvfp4 (NVIDIA). Engines: vLLM, SGLang.
Download (Hugging Face)Weights on Hugging Face ↗huggingface.co ↗huggingface.co ↗huggingface.co ↗
Every test result
Every result
All the numbers,with where they came from.
Every result we've found for this model, with who measured it and a link to where we read it.
All raw rows (vendor and independent kept separate), with setting label, source, date, harness and notes.
| NotesNotes | ||||||
|---|---|---|---|---|---|---|
| AA-LCR | 80.3% | MaxDeepSeek V4 Pro 0813 (Max) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA-LCR accuracy, AA-run (long multi-document reasoning) |
| AA-Omniscience Accuracy | 49.1% | MaxDeepSeek V4 Pro 0813 (Max) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | |
| AA-Omniscience Hallucination Rate | 94.8% | MaxDeepSeek V4 Pro 0813 (Max) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | share of non-correct answers that were wrong instead of abstaining; = 1 - AA 'omniscienceNonHallucination' field (equals breakdown.hallucinationRate on the eval page) |
| AA-Omniscience Index | 0.8 | MaxDeepSeek V4 Pro 0813 (Max) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA-Omniscience Index (-100..100): correct minus incorrect, abstentions not penalised |
| Agents' Last Exam | 25.7% | Maxreasoning_effort=max | Maker's own figureVendor-reportedDeepSeek ↗ | 13 Aug 2026 | — | temperature=1.0, top_p=0.95 |
| Artificial Analysis Coding Agent Index | 43.1 | MaxDeepSeek V4 Pro 0813 (max) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | Codex | agent Codex; components: DeepSWE v1.1 57.2, SWE-Atlas-QnA 61.8, Terminal-Bench v4 10.1; avg cost $0.24/task; avg wall time 40 min/task |
| Artificial Analysis Intelligence Index | 36 | MaxDeepSeek V4 Pro 0813 (Max) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug deepseek-v4-pro; list price $1.32/3.96 per 1M in/out; cost to run AA Intelligence Index $0.67/task |
| Artificial Analysis output speed | 87 tok/s | MaxDeepSeek V4 Pro 0813 (Max) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | tokens/sec (median output speed, first-party API); TTFT 1.7s; list price $1.32/3.96 per 1M in/out |
| AutomationBench | 31.8% | Maxreasoning_effort=max | Maker's own figureVendor-reportedDeepSeek ↗ | 13 Aug 2026 | — | AutomationBench (Public); temperature=1.0, top_p=0.95 |
| Codeforces Elo | 3348 | MaxMax reasoning effort | Maker's own figureVendor-reportedDeepSeek ↗ | 10 Sep 2026 | — | Column 'DS-V4-Pro' (= 0813 GA; TB2.1 87.9 matches) in the V4.1-Flash card comparison |
| CyberGym | 83.3% | Maxreasoning_effort=max | Maker's own figureVendor-reportedDeepSeek ↗ | 13 Aug 2026 | DeepSeek Harness (minimal mode) | temperature=1.0, top_p=0.95 |
| DeepSWE | 57.2% | MaxDeepSeek V4 Pro 0813 (max) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | Codex | AA Coding Agent Index component; pass@1 avg of 3 attempts |
| DeepSWE | 62.7% | Maxreasoning_effort=max | Maker's own figureVendor-reportedDeepSeek ↗ | 13 Aug 2026 | DeepSeek Harness (minimal mode) | DeepSWE (version not stated in this card); temperature=1.0, top_p=0.95 |
| GPQA Diamond | 92.8% | MaxDeepSeek V4 Pro 0813 (Max) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA-run evaluation (AA slug deepseek-v4-pro) |
| GPQA Diamond | 92.4% | Maxreasoning_effort=max | Independent testIndependentVals.ai ↗ | 1 Sep 2026 | — | ±2.02 stderr; $0.066/test |
| GPQA Diamond | 92.4% | MaxMax reasoning effort | Maker's own figureVendor-reportedDeepSeek ↗ | 10 Sep 2026 | — | Column 'DS-V4-Pro' (= 0813 GA; TB2.1 87.9 matches) in the V4.1-Flash card comparison |
| Humanity's Last Exam | 41.0% | MaxDeepSeek V4 Pro 0813 (Max) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA-run evaluation (AA slug deepseek-v4-pro) |
| Humanity's Last Exam | 42.7% | Maxreasoning_effort=max | Maker's own figureVendor-reportedDeepSeek ↗ | 13 Aug 2026 | — | HLE without tools; temperature=1.0, top_p=0.95 |
| Humanity's Last Exam (with tools) | 60.0% | Maxreasoning_effort=max | Maker's own figureVendor-reportedDeepSeek ↗ | 13 Aug 2026 | — | temperature=1.0, top_p=0.95 |
| IOI (Vals) | 51.6% | Maxreasoning_effort=max | Independent testIndependentVals.ai ↗ | 29 Sep 2026 | — | ±2.4 stderr; $2.152/test |
| LegalBench (Vals) | 82.4% | Maxreasoning_effort=max | Independent testIndependentVals.ai ↗ | 29 Sep 2026 | — | vals id deepseek/deepseek-v4-pro-0813; rank 71/149; ±0.437 stderr; $0.004081/test |
| LiveCodeBench | 87.5% | Maxreasoning_effort=max | Independent testIndependentVals.ai ↗ | 1 Sep 2026 | — | ±0.958 stderr; $0.064/test |
| LLM Creative Story-Writing Benchmark (Lech Mazur) | 0.5 | HighDeepSeek V4 Pro (high) | Independent testIndependentLech Mazur (LLM Creative Story-Writing Benchmark) ↗ | 28 Sep 2026 | — | rank 21/56; Thurstone comparison score (centered at 0); est. win chance 58%; 95% bootstrap 0.448 to 0.651 |
| LMArena Text - Creative Writing | 1445 | Highdeepseek-v4-pro-high-20260813 | Independent testIndependentLMArena ↗ | 30 Sep 2026 | — | rank 51 (rank range 20-80); 95% CI 1430.8-1458.6; 2000 votes; style-controlled |
| LMArena Text - Expert | 1484 | Highdeepseek-v4-pro-high-20260813 | Independent testIndependentLMArena ↗ | 30 Sep 2026 | — | rank 63 (rank range 22-111); 95% CI 1465.5-1502.3; 1021 votes; style-controlled |
| LMArena Text - Hard Prompts | 1483 | Highdeepseek-v4-pro-high-20260813 | Independent testIndependentLMArena ↗ | 30 Sep 2026 | — | rank 58 (rank range 37-80); 95% CI 1475.2-1490.8; 6548 votes; style-controlled |
| LMArena Text - Instruction Following | 1461 | Highdeepseek-v4-pro-high-20260813 | Independent testIndependentLMArena ↗ | 30 Sep 2026 | — | rank 46 (rank range 21-74); 95% CI 1451.0-1471.6; 3481 votes; style-controlled |
| LMArena Text - Longer Query | 1475 | Highdeepseek-v4-pro-high-20260813 | Independent testIndependentLMArena ↗ | 30 Sep 2026 | — | rank 52 (rank range 22-75); 95% CI 1465.9-1484.5; 4582 votes; style-controlled |
| LMArena Text - Multi-Turn | 1472 | Highdeepseek-v4-pro-high-20260813 | Independent testIndependentLMArena ↗ | 30 Sep 2026 | — | rank 54 (rank range 15-90); 95% CI 1456.5-1487.2; 1573 votes; style-controlled |
| LMArena Text - Non-English | 1447 | Highdeepseek-v4-pro-high-20260813 | Independent testIndependentLMArena ↗ | 30 Sep 2026 | — | rank 57 (rank range 41-83); 95% CI 1439.2-1455.1; 6054 votes; style-controlled |
| LMArena Text - Occupational: Business, Management & Finance | 1468 | Highdeepseek-v4-pro-high-20260813 | Independent testIndependentLMArena ↗ | 30 Sep 2026 | — | rank 41 (rank range 12-84); 95% CI 1453.8-1481.8; 1816 votes; style-controlled |
| LMArena Text - Occupational: Legal & Government | 1483 | Highdeepseek-v4-pro-high-20260813 | Independent testIndependentLMArena ↗ | 30 Sep 2026 | — | rank 34 (rank range 4-96); 95% CI 1462.2-1503.2; 831 votes; style-controlled |
| LMArena Text - Occupational: Writing, Literature & Language | 1454 | Highdeepseek-v4-pro-high-20260813 | Independent testIndependentLMArena ↗ | 30 Sep 2026 | — | rank 43 (rank range 20-72); 95% CI 1442.2-1466.1; 2645 votes; style-controlled |
| MathArena Apex | 65.3% | MaxMax reasoning effort | Maker's own figureVendor-reportedDeepSeek ↗ | 10 Sep 2026 | — | Column 'DS-V4-Pro' (= 0813 GA; TB2.1 87.9 matches) in the V4.1-Flash card comparison |
| NL2Repo-Bench | 61.5% | Maxreasoning_effort=max | Maker's own figureVendor-reportedDeepSeek ↗ | 13 Aug 2026 | DeepSeek Harness (minimal mode) | temperature=1.0, top_p=0.95 |
| ProgramBench (Almost Solved) | 15.5% | MaxMax reasoning effort | Maker's own figureVendor-reportedDeepSeek ↗ | 10 Sep 2026 | DeepSeek Harness (minimal mode) | Column 'DS-V4-Pro' (= 0813 GA; TB2.1 87.9 matches) in the V4.1-Flash card comparison |
| ProgramBench (fully resolved) | 0.0% | Maxreasoning_effort=max | Independent testIndependentVals.ai ↗ | 29 Sep 2026 | mini-SWE-agent | ±0 stderr; $0.432/test; strict fully-resolved rate |
| SciCode | 51.0% | MaxDeepSeek V4 Pro 0813 (Max) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA-run evaluation (AA slug deepseek-v4-pro) |
| SimpleQA Verified | 52.9% | Maxdeepseek-v4-pro-0813_max | Independent testIndependentEpoch AI ↗ | 27 Aug 2026 | — | Epoch-run (no tools); ±1.58 stderr |
| SkillsBench | 53.8% | Maxreasoning_effort=max | Independent testIndependentVals.ai ↗ | 27 Sep 2026 | OpenHands | ±4.49 stderr; $0.375/test |
| SWE Atlas - Codebase QnA | 61.8% | MaxDeepSeek V4 Pro 0813 (max) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | Codex | AA Coding Agent Index component; pass@1 avg of 3 attempts |
| SWE-bench Verified | 96.4% | Maxreasoning_effort=max | Independent testIndependentVals.ai ↗ | 1 Sep 2026 | bash-only (single bash tool) agent | ±0.834 stderr; $0.103/test; Vals archived SWE-bench Verified on 2026-09-01 (saturated) |
| Terminal-Bench 2.1 | 78.7% | MaxDeepSeek V4 Pro 0813 (Max) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA-run evaluation (AA slug deepseek-v4-pro) |
| Terminal-Bench 2.1 | 87.9% | Maxreasoning_effort=max | Maker's own figureVendor-reportedDeepSeek ↗ | 13 Aug 2026 | DeepSeek Harness (minimal mode) | temperature=1.0, top_p=0.95 |
| Terminal-Bench 3.0 | 11.8% | MaxMax reasoning effort | Maker's own figureVendor-reportedDeepSeek ↗ | 10 Sep 2026 | DeepSeek Harness (minimal mode) | Column 'DS-V4-Pro' (= 0813 GA; TB2.1 87.9 matches) in the V4.1-Flash card comparison |
| Terminal-Bench 4.0 | 14.1% | Maxreasoning_effort=max | Independent testIndependentVals.ai ↗ | 29 Sep 2026 | mini-SWE-agent | ±2.02 stderr; $3.314/test |
| Terminal-Bench 4.0 | 14.1% | MaxDeepSeek V4 Pro 0813 (Max) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA-run evaluation (AA slug deepseek-v4-pro) |
| Terminal-Bench 4.0 | 10.1% | MaxDeepSeek V4 Pro 0813 (max) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | Codex | AA Coding Agent Index component run in the vendor agent harness; pass@1 avg of 3 attempts |
| Terminal-Bench 4.0 | 12.4% | MaxMax reasoning effort | Maker's own figureVendor-reportedDeepSeek ↗ | 10 Sep 2026 | DeepSeek Harness (minimal mode) | Column 'DS-V4-Pro' (= 0813 GA; TB2.1 87.9 matches) in the V4.1-Flash card comparison |
| Toolathlon-Verified | 74.1% | Maxreasoning_effort=max | Maker's own figureVendor-reportedDeepSeek ↗ | 13 Aug 2026 | — | temperature=1.0, top_p=0.95 |
| Vals Code Migration | 41.5% | Maxreasoning_effort=max | Independent testIndependentVals.ai ↗ | 29 Sep 2026 | — | ±4.301 stderr; $18.581/test |
| Vals CorpFin v2 | 65.4% | Maxreasoning_effort=max | Independent testIndependentVals.ai ↗ | 12 Aug 2026 | — | vals id deepseek/deepseek-v4-pro-0813; rank 29/134; ±0.937 stderr; $0.002183/test |
| Vals Legal Research Bench | 40.9% | Maxreasoning_effort=max | Independent testIndependentVals.ai ↗ | 29 Sep 2026 | — | vals id deepseek/deepseek-v4-pro-0813; rank 24/72; ±3.417 stderr; $1.088017/test |
| Vals TaxEval v2 | 73.1% | Maxreasoning_effort=max | Independent testIndependentVals.ai ↗ | 1 Sep 2026 | — | vals id deepseek/deepseek-v4-pro-0813; rank 55/145; ±0.872 stderr; $0.045816/test |
| Vals Vibe Code Bench | 82.3% | Maxreasoning_effort=max | Independent testIndependentVals.ai ↗ | 29 Sep 2026 | OpenHands | ±3.148 stderr; $0.356/test |