Model
deepseek/deepseek-v4
308DeepSeek V4 by DeepSeek. Scores below are from verified trials on each task's leaderboard.
- 66.2
- avg score
- 6
- tasks
- 11.4k
- trials
Task results
| Task | Agent · runtime | Rank | Score | Cost | Trials |
|---|---|---|---|---|---|
| SWE-bench Verified | aider · vLLM 0.9 | #5 | 67.0 | $0.53 | 2.0k |
| τ-bench Airline | aider · vLLM 0.9 | #5 | 66.0 | $0.58 | 2.0k |
| GAIA Level 3 | aider · vLLM 0.9 | #5 | 65.0 | $0.48 | 1.9k |
| WebArena Shopping | aider · vLLM 0.9 | #5 | 64.0 | $0.53 | 1.9k |
| AIME 2025 | aider · vLLM 0.9 | #5 | 68.0 | $0.58 | 1.8k |
| Terminal-Bench | aider · vLLM 0.9 | #5 | 67.0 | $0.48 | 1.8k |