TAU3-Bench

TAU3-Bench - agentic tool use benchmark. Uses official user model (gpt-5.2, low reasoning effort) + default BM25 retrieval.

Category

General Agents

Max Score

100.0

Score Type

percent

Active

Yes

Model Scores

Scores for TAU3-Bench

Model Score Date Verified
Qwen3.6-35B-A3B
97.49%
16.04.2026 Verified
Gemma4-31B-it
97.93%
16.04.2026 Unverified
Qwen3.5-35B-A3B
100.00%
16.04.2026 Unverified
Gemma4-26B-A4B
85.38%
16.04.2026 Unverified
Qwen3.5-27B
99.26%
16.04.2026 Unverified
Spark-X2.5-4B
43.13%
— Verified
Spark-X2.5-1.7B
27.92%
— Verified
Qwen3.5-9B
11.96%
— Verified
Qwen3.5-4B
8.12%
— Verified
Qwen3.5-2B
4.28%
— Verified
Gemma4-12B
17.87%
— Verified
Gemma4-E4B
13.15%
— Verified
Gemma4-E2B
11.23%
— Verified
MiniCPM5-2B
28.95%
— Verified
granite-4.2-3B
6.50%
— Verified
Nemotron-3-Nano-4B
1.20
— Verified
LFM2.5-8B-A1B
3.25%
— Verified
LFM2.5-2.6B
8.86%
— Verified