STT-Arena - Medium: leaderboard
Metric: Pass@1 (%) on the 67 medium tasks; a task passes when every rule-based check function holds in the final state; 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 August 2028. 23 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.6 | 31.34 |
| 2 | Gemini 3.1 Pro (Preview) | 30.35 |
| 3 | Claude Sonnet 4.6 | 29.85 |
| 4 | GPT-5.4 | 29.85 |
| 5 | DeepSeek V3.2 | 27.86 |
| 6 | Qwen 3.6 Plus | 24.38 |
| 7 | GLM-5.1 | 23.38 |
| 8 | GLM-5 | 21.39 |
| 9 | DeepSeek V4 Pro | 20.4 |
| 10 | MiniMax-M2.7 | 15.92 |
| 11 | MiniMax-M2.5 | 15.92 |
| 12 | Gemini 2.5 Flash | 13.43 |
| 13 | Kimi K2.5 | 13.43 |
| 14 | Qwen 3 8B | 13.43 |
| 15 | GPT-5.4 Mini | 13.43 |
Interactive version: theaggregate.ai/benchmark?slug=stt-arena-medium · How It Works · Data refreshed daily, snapshot 2026-10-07.