DeepSeek V3 (March 2025): benchmark results
March 2025 DeepSeek V3 snapshot. Provider: DeepSeek. Released 2025-03-24. Access: Open.
Aggregate Rating 1624 ± 25, rank #795 of 10916 rated models, from 17 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| SEAL - Agentic Tool Use (Enterprise) | 64.23 | Score | 69.7 |
| SEAL - Agentic Tool Use (Chat) | 54.31 | Score | 54.5 |
| Epoch AI - ECI | 135.23 | ECI Score | 45.6 |
| HAL TAU-bench Airline | 44 | Accuracy (%) | 42.9 |
| HAL SciCode | 3.08 | Accuracy (%) | 40 |
| HAL Online Mind2Web | 32.33 | Accuracy (%) | 36.4 |
| HAL USACO | 39.09 | Accuracy (%) | 36.4 |
| HAL CORE-Bench Hard | 17.78 | Accuracy (%) | 25 |
| HAL GAIA Level 2 | 31.97 | Accuracy (%) | 12.5 |
| HAL SWE-bench Verified Mini | 24 | Score (%) | 11.8 |
| SEAL - Humanity's Last Exam (Text Only) | 4.55 | Score | 10.2 |
| HAL AssistantBench | 2.03 | Accuracy (%) | 7.1 |
Interactive version: theaggregate.ai/model?slug=deepseek-v3-march-2025 · How It Works · Data refreshed daily, snapshot 2026-11-08.