List your company

Agentic Compositional Generalization on SkillBench

8.69Skill Depth (k=2)GPT-5.2
3.93725.17116.4057.6389Jan 7, 2026

Evaluation results

MethodLinks
GPT-5.2Model Category=Advanced Agentic Models2026.018.698.728.848.678.668.178.62
DeepSeek-v3.2Model Category=Advanced Agentic Models2026.018.428.438.088.168.1988.21
Gemini-3-ProModel Category=Advanced Agentic Models2026.017.877.897.897.77.596.817.62
Qwen2.5-32BModel Category=Instruct Models2026.015.55.164.895.255.295.165.21
Qwen2.5-14BModel Category=Instruct Models2026.015.395.2255.045.134.945.12
Qwen2.5-7BModel Category=Instruct Models2026.015.245.054.814.915.244.975.04
SearchR1Model Category=Tool-Integrated Agentic Models2026.014.734.364.444.534.824.324.54
ReSearchModel Category=Tool-Integrated Agentic Models2026.014.724.464.064.374.414.354.4
AutoTIRModel Category=Tool-Integrated Agentic Models2026.014.124.214.14.034.074.024.09