DeepSeek
DeepSeek-V3 benchmark results
8 results on 7 benchmarks from 4 sources.
Open weights Released: Dec 26, 2024 Updated Oct 22, 2026
Also listed as: deepseek-v3 · DeepSeek v3 · DeepSeek-V3 Base · deepseek-chat · Deepseek-V3
All benchmark results
| Benchmarks | Score | Rank | Setting | Source type | Date |
|---|---|---|---|---|---|
| LMArena Text | 1358 | 206 | Overall, style control on; 21770 votes; 95% CI 1354–1363“deepseek-v3” | Official leaderboard | Oct 8, 2026 |
| LiveCodeBench | 14.9% | 22 | Code generation, 182 problems dated 2025-01-01…2025-04-07 | Official leaderboard | Apr 7, 2025 |
| GPQA Diamond | 56.5% | 188 | Epoch AI run, accuracy | Independent run | Jan 27, 2025 |
| MMLU | 87.1% | 2 | Exact match; 5-shot | Paper / tech report via Epoch AI | Dec 26, 2024 |
| MMLU | 87.2% | 2 | Exact match; 5-shot | Official leaderboard via Epoch AI | Dec 26, 2024 |
| MMLU-Pro | 75.9% | 84 | Overall accuracy; data source: Self-Reported; 671B“Deepseek-V3” | Vendor-reported | Mar 11, 2026 |
| AIME (OTIS Mock AIME 2024–2025) | 15.8% | 212 | Epoch AI run, OTIS Mock AIME 2024–2025 | Independent run | Feb 25, 2025 |
| MATH Level 5 | 64.8% | 48 | Epoch AI run, accuracy | Independent run | Jan 27, 2025 |
FAQ
Where do these DeepSeek-V3 scores come from?
From Epoch AI Benchmarking Hub, LiveCodeBench leaderboard, LMArena leaderboard dataset, and MMLU-Pro leaderboard (TIGER-Lab). Each row links to the original leaderboard, dataset or report, and vendor-reported numbers are labelled.