STT-Arena - Easy: leaderboard

Easy tasks, a single conflict that needs one corrective action, of STT-Arena (2026), a tool-use benchmark in executable stateful environments whose conditions change in time and space during the task, with a simulated user. Pass@1 in percent (every rule-based check must hold in the final state), mean of three runs; higher is better.

Metric: Pass@1 (%) on the 71 easy tasks (a single isolated conflict needing one corrective action); 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 January 2028. 21 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Claude Opus 4.641.78#114
2GPT-5.439.91#138
3Gemini 3.1 Pro (Preview)37.56#102
4DeepSeek V3.236.62#425
5Claude Sonnet 4.635.21#179
6Qwen 3.6 Plus35.21#245
7GLM-5.135.21#200
8DeepSeek V4 Pro32.86#201
9GLM-530.05#276
10STT-Agent-4B (SFT+RL)29.11#324
11Gemini 2.5 Flash28.17#466
12Qwen 3.5 9B27.23#838
13MiniMax-M2.526.76#600
14STT-Agent-4B (SFT)26.76#424
15MiniMax-M2.726.29#539

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=stt-arena-easy · How It Works · Data refreshed daily, snapshot 2026-11-22.