Agent on τ2-Bench benchmark leaderboard · SOTA2 Research

Agent on τ2-Bench

85.4AccuracyGemini-3-Pro

Evaluation Results

Method	Links
Gemini-3-Pro2026.03		85.4
Intern-S1-Pro2026.03		80.9
Kimi-K2.52026.03		76.8
GPT-5.22026.03		76.6
JT-Safe-V2-35B2026.05		70
LongCat-Flash Exp-Chat2025.12		69.5
GLM 4.62025.12		69.1
LongCat-Flash Chat2025.12		68.8
Lowest Centroid2026.04		65
DeepSeek V3.22025.12		64
Greedy Decoding2026.04		58
Qwen3-VL-235B-Thinking2026.03		57.4
Self-Certainty2026.04		57
Pass@12026.04		54.9
Tail Confidence2026.04		53
Bottom Window2026.04		49
SOTA with Equivalent Parameters2026.05		48.1
Lowest Centroid2026.04		35
Greedy Decoding2026.04		33
Lowest Centroid2026.04		33
Greedy Decoding2026.04		29
Pass@12026.04		26.5
Pass@12026.04		25.9
Bottom Window2026.04		23
Self-Certainty2026.04		22
Tail Confidence2026.04		21
Tail Confidence2026.04		21
Bottom Window2026.04		21
Self-Certainty2026.04		18
Lowest Centroid2026.04		18
Pass@12026.04		12.2
Lowest Centroid2026.04		12
Greedy Decoding2026.04		12
Pass@12026.04		9.5
Self-Certainty2026.04		9
Bottom Window2026.04		9
Bottom Window2026.04		8
Tail Confidence2026.04		8
Greedy Decoding2026.04		7
Self-Certainty2026.04		7
Tail Confidence2026.04		4