Inference
Sierra Research

τ³-Bench Tool-Agent-User Evaluation

τ³-bench is the current evolution of Sierra's tool-agent-user framework, adding corrected task releases and newer knowledge and voice evaluation modes alongside airline, retail, and telecom.

ModelsModels with a result.18
Top resultBest result on this test.96.0%Mercury 2.5
Top-3 spreadPoints from first to third.23.1 pts
YearYear of release.2026

Result and price

020406080100$0.1$0.3$1$3$10$30RESULTOUTPUT PRICE PER 1M TOKENS · LOG SCALE

Results

18 results
#ModelResultThe score from the source.
01Mercury 2.596.0%
02Mistral Medium 3.5 128B91.4%
03MiMo-V2.5-Pro72.9%
04Nemotron 3 Ultra70.9%
05Qwen3.6 Plus70.7%
06GLM-5.170.6%
07Claude Opus 4.570.2%
08Qwen3.5 397B68.4%
09Qwen3.6-35B-A3B67.2%
10Pokee-Isaac 28B66.2%
11Kimi K2.565.7%
12GLM-565.6%
13Granite 4.2 30B62.0%
14Granite 4.2 8B58.1%
15Granite 4.2 3B45.8%
16Atria Dawn Preview41.2%
17Nemotron 3.5 Lightning 30B A3B NVFP49.5%
18LFM2.5-2.6B5.7%