Every run: the agent's full trajectory, its submissions over the 20-hour budget, and the final score.
| # | Model | Reward | Cost | In tok | Out tok | Time | Steps | Submits |
|---|---|---|---|---|---|---|---|---|
| 1 | DeepSeek V4 Flash Vision Exp | 0.0671 | $3.79 | 218.1M | 451k | 16h 2m | 591 | 4 |
| 2 | DeepSeek V4 Flash Vision Exp | 0.0421 | $4.04 | 235.7M | 470k | 19h 34m | 605 | 5 |
| 3 | DeepSeek V4 Flash Vision Exp | 0.0178 | $3.85 | 221.0M | 459k | 11h 59m | 555 | 7 |
| 4 | DeepSeek V4 Flash Vision Exp | 0.0114 | $4.73 | 279.3M | 524k | 19h 31m | 681 | 4 |
| 5 | DeepSeek V4 Flash Vision Exp | 0.0000† | $2.30 | 127.9M | 331k | 19h 43m | 454 | 3 |
† score adjusted by manual review; the trace's Judge tab carries the measured score and the rationale.