STT-Arena - Impossible: leaderboard
Metric: Pass@1 (%) on the 30 impossible tasks (no valid completion path exists); a task passes when a Qwen3.5-397B judge finds that the model both recognized the infeasibility and told the user; the model acts in an executable stateful environment whose state changes under scheduled spatio-temporal triggers, with a passive Qwen3.5-397B user simulator and at most 50 turns; mean of three runs at temperature 0.7; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 23 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek V3.2 | 63.33 |
| 2 | Claude Sonnet 4.6 | 58.89 |
| 3 | Qwen 3.6 Plus | 54.44 |
| 4 | Gemini 3.1 Pro (Preview) | 52.22 |
| 5 | GPT-5.4 | 52.22 |
| 6 | Claude Opus 4.6 | 52.22 |
| 7 | DeepSeek V4 Pro | 45.56 |
| 8 | GLM-5 | 43.33 |
| 9 | MiniMax-M2.5 | 42.22 |
| 10 | GLM-5.1 | 41.11 |
| 11 | Kimi K2.5 | 37.78 |
| 12 | MiniMax-M2.7 | 36.67 |
| 13 | Qwen 3.5 9B | 32.22 |
| 14 | Gemini 2.5 Flash | 31.11 |
| 15 | Llama 3.3 70B | 30 |
Interactive version: theaggregate.ai/benchmark?slug=stt-arena-impossible · How It Works · Data refreshed daily, snapshot 2026-10-07.