SOTA Agent on τ2-Bench and PapersWithCode

85.4Accuracy

Gemini-3-Pro

Updated 8d ago

Evaluation Results

Method	Links
Gemini-3-Pro 2026.03		85.4
Intern-S1-Pro 2026.03		80.9
Kimi-K2.5 2026.03		76.8
GPT-5.2 2026.03		76.6
JT-Safe-V2-35B 2026.05		70
LongCat-Flash Exp-Chat 2025.12		69.5
GLM 4.6 2025.12		69.1
LongCat-Flash Chat 2025.12		68.8
Lowest Centroid 2026.04		65
DeepSeek V3.2 2025.12		64
Greedy Decoding 2026.04		58
Qwen3-VL-235B-Thinking 2026.03		57.4
Self-Certainty 2026.04		57
Pass@1 2026.04		54.9
Tail Confidence 2026.04		53
Bottom Window 2026.04		49
SOTA with Equivalent Parameters 2026.05		48.1
Lowest Centroid 2026.04		35
Greedy Decoding 2026.04		33
Lowest Centroid 2026.04		33
Greedy Decoding 2026.04		29
Pass@1 2026.04		26.5
Pass@1 2026.04		25.9
Bottom Window 2026.04		23
Self-Certainty 2026.04		22
Tail Confidence 2026.04		21
Tail Confidence 2026.04		21
Bottom Window 2026.04		21
Self-Certainty 2026.04		18
Lowest Centroid 2026.04		18
Pass@1 2026.04		12.2
Lowest Centroid 2026.04		12
Greedy Decoding 2026.04		12
Pass@1 2026.04		9.5
Self-Certainty 2026.04		9
Bottom Window 2026.04		9
Bottom Window 2026.04		8
Tail Confidence 2026.04		8
Greedy Decoding 2026.04		7
Self-Certainty 2026.04		7
Tail Confidence 2026.04		4