Multi-agent coordination on CollabEscape (test)
80Success RateGPT-4-Turbo
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-4-Turbo2026.06 | 80 | 4.6 | 4 | 520 | |
| MA-Debatemodel_size=235B2026.06 | 80 | 4.4 | 8 | 480 | |
| DICE-FTagents=3×8B2026.06 | 80 | 3.9 | 19 | 365 | |
| DICE-FTagents=3×32B2026.06 | 80 | 3.4 | 29 | 445 | |
| GPT-o32026.06 | 70 | 4.2 | 13 | 485 | |
| Qwen3-235B-Inst2026.06 | 70 | 4.9 | 2 | 465 | |
| MADagents=3×8B2026.06 | 70 | 4.8 | — | 395 | |
| MA-Debatemodel_size=32B2026.06 | 70 | 4.5 | 6 | 425 | |
| MA-FTmodel_size=32B2026.06 | 70 | 4.3 | 10 | 435 | |
| Debate+Judgebackbone=GPT-42026.06 | 70 | 4.7 | 2 | 545 | |
| DICE-PCagents=3×8B2026.06 | 70 | 4.1 | 15 | 380 | |
| Qwen3-32B-Inst2026.06 | 60 | 5.3 | 10 | 410 | |
| rStar2026.06 | 60 | 5.1 | 6 | 470 |