DeepSeek-V4-Pro (High, 20260813)
Made by DeepSeek, released 13 Aug 2026. Epoch AI lists its access as “Open weights (unrestricted)”. Appears solid at knowledge and writing among the models listed here.
In short
What it does well
- Knowledge: Strong (19th of 43 listed models)
- Writing: Strong (37th of 77 listed models)
Where it falls short
- Coding: Middle (52nd of 77 listed models)
- Maths: Middle (27th of 43 listed models)
What we don't know yet
- Fixing real software bugs: Not enough data
The four ratings
- Top tier
- Strong
- Middle
- Behind the leaders
- Not enough data
Ratings compare only the 77 models listed here, in quarters, best first. “Not enough data” means no result yet. It is never a low score.
- Writing
- StrongHow often people preferred its answers to writing prompts. 1,440 on Arena's scale · Source for DeepSeek-V4-Pro (High, 20260813) writing score (lmarena)
- Coding
- MiddleHow often people preferred its answers to coding questions. 1,469 on Arena's scale · Source for DeepSeek-V4-Pro (High, 20260813) coding score (lmarena)
- Maths
- MiddleHow many very hard maths problems it solved. 64.6% · Source for DeepSeek-V4-Pro (High, 20260813) maths score (epoch)
- Knowledge
- StrongHow many short factual questions it answered correctly. 52.9% · Source for DeepSeek-V4-Pro (High, 20260813) knowledge score (epoch)
Evidence completeness: Broad evidence, 12 of 13 tests have a result.
Where it is relatively stronger or weaker
This compares each kind of request with the model's own overall position among the 77 listed models, using Arena's votes. It is about performance on different languages, kinds of request and topics, not about political or social bias.
Relatively stronger
No category stands out as clearly stronger than its overall position.
Relatively weaker
No category stands out as clearly weaker than its overall position.
Show every category we can see
| Category | Position | Arena score | Votes | Source |
|---|---|---|---|---|
| Expert-level prompts Kind of request | 56th of 77 | 1,459 95% range 1,441 to 1,478 | 1,032 | Source for Expert-level prompts |
| Hard prompts Kind of request | 49th of 77 | 1,458 95% range 1,450 to 1,466 | 6,522 | Source for Hard prompts |
| Long prompts Kind of request | 43rd of 77 | 1,458 95% range 1,448 to 1,467 | 4,558 | Source for Long prompts |
| Longer conversations Kind of request | 46th of 77 | 1,454 95% range 1,439 to 1,469 | 1,564 | Source for Longer conversations |
| Chinese Language | 56th of 76 | 1,482 95% range 1,460 to 1,505 | 707 | Source for Chinese |
| English prompts Language | 44th of 77 | 1,458 95% range 1,448 to 1,468 | 3,761 | Source for English prompts |
| French Language | 62nd of 72 | 1,447 95% range 1,414 to 1,480 | 316 | Source for French |
| Non-English prompts Language | 59th of 77 | 1,431 95% range 1,423 to 1,439 | 6,026 | Source for Non-English prompts |
| Russian Language | 50th of 77 | 1,447 95% range 1,431 to 1,464 | 1,219 | Source for Russian |
| Entertainment, sport and media Topic | 48th of 77 | 1,424 95% range 1,412 to 1,436 | 2,447 | Source for Entertainment, sport and media |
| Law and government Topic | 44th of 76 | 1,459 95% range 1,439 to 1,480 | 825 | Source for Law and government |
| Medicine and health Topic | 54th of 75 | 1,446 95% range 1,423 to 1,469 | 626 | Source for Medicine and health |
| Science Topic | 51st of 77 | 1,458 95% range 1,443 to 1,474 | 1,525 | Source for Science |
| Software and IT Topic | 59th of 77 | 1,461 95% range 1,452 to 1,471 | 4,093 | Source for Software and IT |
Bias: what we can and can't see
Our sources do not measure political lean, stereotyping, flattery (agreeing with you too readily) or safety behaviour, so we cannot say how any model does on those. What we can show is above: where a model is relatively stronger or weaker across languages, kinds of request and topics. Also, people voting in Arena can favour answer style, such as length and formatting, and our numbers do not adjust for that. See what AI bias means and what these tests can tell you.
Best-supported use cases
- Research and checking facts: best open-weights option.
Where an alternative might be more suitable
- Writing and creating content: rated Strong on writing votes. Compare it with Claude Opus 5.5, the best-supported choice for this task.
- Coding and building things: rated Middle on coding votes. Compare it with Claude Opus 5.5, the best-supported choice for this task.
- Homework and maths: rated Middle on hard maths problems. Compare it with GPT-6.1 Sol, the best-supported choice for this task.
- Business emails and everyday work: rated Middle on business-task votes. Compare it with Gemini 4 Argon, the best-supported choice for this task.
- Research and checking facts: rated Strong on short factual questions. Compare it with Gemini 4 Argon, the best-supported choice for this task.
What it is
- Maker
- DeepSeek
- Release date (Epoch AI)
- 13 Aug 2026
- Access, as listed by Epoch AI
- Open weights (unrestricted)
- Licence, as listed by Arena
- MIT
- Maker's country (Epoch AI)
- China
- Last observation in our data
- 28 Oct 2026
- Data last changed
- 29 Oct 2026, 14:36 UTC
We have not verified an official product page for this model, so we do not link to one.
How you can use it
Open weights. Arena lists its licence as MIT. In plain terms, the model's files are published under a licence that lets others run it, so it can be offered by more than one company, sometimes at lower cost. That does not guarantee that any particular app is free. Open licences differ, and some restrict commercial use, so check the licence terms before building a business on it.
How you can reach it (an app, a website, a developer API or a third-party tool) and what that costs are not covered by our benchmark data. Where we could verify prices from the maker's own pages, they are shown under Price. Epoch AI lists access as “Open weights (unrestricted)”. See models, apps and harnesses and how AI pricing works.
Price
Price not independently verified
Developer (API) price, USD ($ as shown) per 1 million tokens, billed by use. Checked 29 Oct 2026. Maker's page
| What is counted | Price per 1M tokens |
|---|---|
| Input, new text | 0.66 off-peak, 1.32 peak |
| Input, repeated text the system has seen | 0.022 off-peak, 0.044 peak |
| Output, what the AI writes | 1.98 off-peak, 3.96 peak |
Off-peak rates are half the peak rates. Peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, excluding Chinese public holidays. All other hours are off-peak. This is what developers pay. It is not a subscription price, and it does not say what any app costs. New to this? Read how AI pricing works.
See DeepSeek API models and pricing (checked 29 Oct 2026). Page opened and read. It gives developer (API) prices per 1M tokens, shown on model pages. The chat app (chat.deepseek.com) goes straight to a sign-in page and we did not sign in, so we make no statement about whether it is free.
We never infer subscription prices from API prices. New to “tokens” and “per million”? See how AI pricing works.
Good to remember
- Benchmarks are samples of tasks. None proves performance in every situation.
- Arena scores come from people choosing between answers. That measures preference, not correctness.
- Ratings are relative to the 77 models listed here. A different set would change them.
- Sources test models at different settings. Each result shows the setting used, and we do not adjust it.
The evidence, for the curious
Show all 13 tests with settings and sources
| Test | Result | Rating | Setting used | Source identifier | Observed | Links |
|---|---|---|---|---|---|---|
| Arena text: overall preference Arena 'text' leaderboard (Bradley-Terry rating, no style control) |
1,446 Arena score 95% range 1,439 to 1,453, 9,787 votes |
Middle | effort: High, 20260813, bradley-terry rating | DeepSeek-V4-Pro (High, 20260813) lmarena · text/latest |
28 Oct 2026 | Source for DeepSeek-V4-Pro (High, 20260813) Arena text: overall preference score (lmarena) |
| Arena text: creative writing Arena 'text' leaderboard, category creative_writing |
1,440 Arena score 95% range 1,426 to 1,453, 1,976 votes |
Strong | effort: High, 20260813, bradley-terry rating | DeepSeek-V4-Pro (High, 20260813) lmarena · text/latest |
28 Oct 2026 | Source for DeepSeek-V4-Pro (High, 20260813) Arena text: creative writing score (lmarena) |
| Arena text: coding Arena 'text' leaderboard, category coding |
1,469 Arena score 95% range 1,457 to 1,480, 2,809 votes |
Middle | effort: High, 20260813, bradley-terry rating | DeepSeek-V4-Pro (High, 20260813) lmarena · text/latest |
28 Oct 2026 | Source for DeepSeek-V4-Pro (High, 20260813) Arena text: coding score (lmarena) |
| Arena text: maths Arena 'text' leaderboard, category math |
1,442 Arena score 95% range 1,414 to 1,470, 407 votes |
Behind the leaders | effort: High, 20260813, bradley-terry rating | DeepSeek-V4-Pro (High, 20260813) lmarena · text/latest |
28 Oct 2026 | Source for DeepSeek-V4-Pro (High, 20260813) Arena text: maths score (lmarena) |
| Arena text: following instructions Arena 'text' leaderboard, category instruction_following |
1,449 Arena score 95% range 1,438 to 1,459, 3,474 votes |
Middle | effort: High, 20260813, bradley-terry rating | DeepSeek-V4-Pro (High, 20260813) lmarena · text/latest |
28 Oct 2026 | Source for DeepSeek-V4-Pro (High, 20260813) Arena text: following instructions score (lmarena) |
| Arena text: business and management Arena 'text' leaderboard, category industry_business_and_management_and_financial_operations |
1,442 Arena score 95% range 1,428 to 1,456, 1,810 votes |
Middle | effort: High, 20260813, bradley-terry rating | DeepSeek-V4-Pro (High, 20260813) lmarena · text/latest |
28 Oct 2026 | Source for DeepSeek-V4-Pro (High, 20260813) Arena text: business and management score (lmarena) |
| Arena WebDev: building web apps Arena 'webdev' leaderboard (Code Arena), category overall |
1,582 Arena score 95% range 1,572 to 1,591, 4,676 votes |
Strong | effort: High, 20260813, bradley-terry rating | DeepSeek-V4-Pro (High, 20260813) lmarena · webdev/latest |
28 Oct 2026 | Source for DeepSeek-V4-Pro (High, 20260813) Arena WebDev: building web apps score (lmarena) |
| FrontierMath Tiers 1-3 (v2) FrontierMath Tiers 1-3 v2 (private set), benchmark released 2026-06-12 |
64.6% share of problems solved standard error ±2.8 points |
Middle | effort: max, best score across scorers | deepseek-v4-pro-0813_max epoch · frontiermath_tiers_1_3_v2.csv |
19 Aug 2026 | Source for DeepSeek-V4-Pro (High, 20260813) FrontierMath Tiers 1-3 (v2) score (epoch) · Run details for FrontierMath Tiers 1-3 (v2) |
| OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025, benchmark released 2024-12-19 |
98.6% share of problems solved standard error ±1.0 points |
Strong | effort: max, best score across scorers | deepseek-v4-pro-0813_max epoch · otis_mock_aime_2024_2025.csv |
18 Aug 2026 | Source for DeepSeek-V4-Pro (High, 20260813) OTIS Mock AIME 2024-2025 score (epoch) No public run log in the data file. |
| GPQA Diamond GPQA Diamond, benchmark released 2023-11-20 |
91.7% share of questions correct standard error ±1.5 points |
Strong | effort: max, best score across scorers | deepseek-v4-pro-0813_max epoch · gpqa_diamond.csv |
18 Aug 2026 | Source for DeepSeek-V4-Pro (High, 20260813) GPQA Diamond score (epoch) No public run log in the data file. |
| SimpleQA Verified SimpleQA Verified, benchmark released 2025-09-09 |
52.9% share of questions correct standard error ±1.6 points |
Strong | effort: max, best score across scorers | deepseek-v4-pro-0813_max epoch · simpleqa_verified.csv |
27 Aug 2026 | Source for DeepSeek-V4-Pro (High, 20260813) SimpleQA Verified score (epoch) · Run details for SimpleQA Verified |
| SWE-bench Verified (Epoch runs) | Not enough data | |||||
| FrontierCode (reported by Cognition) FrontierCode, benchmark released 2026-06-08, aggregation Mean@5 |
28.5% mean score over 5 attempts no uncertainty figure published |
Not rated (results use different tools) | effort: high, tool: devin-cli, mean@5 | deepseek-v4-pro-0813_high epoch · frontiercode_external.csv |
not provided | Source for DeepSeek-V4-Pro (High, 20260813) FrontierCode (reported by Cognition) score (epoch) · Run details for FrontierCode (reported by Cognition) |
Missing evidence
Not enough data for these tests:
- SWE-bench Verified (Epoch runs)
Missing results are not low scores and do not count against the model.
Where a source gives no public run log, we cannot show the run behind the score. See data sources.
Not sure this is the right one?
Tell us what you need help with and we'll narrow the options down.