tau-benchactiveAgentsTool use
tau-bench: τ-bench
Metric: % Task Success (higher is better)Introduced: 2024
Multi-turn tool-use benchmark for agent behavior in realistic retail and airline customer-service tasks, measuring whether models complete tasks through APIs while following policy constraints.
Models ranked
29
tracked on this benchmark
Score band
95.3 – 61.1
best → lowest tracked
Snapshot trend
-7.71
May 5 → Jun 7 · 13 models
Leaderboard
Tracked models ranked by % Task Success (higher is better).
#Model variant and provenanceRelative to leaderScore
1
Ring-2.6-1T
95.3Version: Tau2-Bench Telecom scenario (accuracy)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
2
Mistral Medium 3.5
91.4Version: τ³-Telecom (agentic tool use) (accuracy)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
3
ByteDance Doubao Seed 2.0 Pro
90.4Version: τ²-Bench Retail (accuracy)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
4
Claude Mythos Preview
89.2Version: τ-benchHarness: Not recordedEvaluator: Not recordedObserved: May 5, 2026Confidence: Not recordedSource
Notes: Confidence: high. DAT-3780 weekly best-of refresh: benchlm.ai TAU-bench snapshot lists Claude Mythos Preview at 89.2%.
5
LFM2.5 8B A1B
88.1Version: τ² Telecom benchmark (accuracy)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
6
Claude Sonnet 4.6
87.5Version: τ-benchHarness: Not recordedEvaluator: Not recordedObserved: Apr 24, 2026Confidence: Not recordedSource
7
Qwen3.5-397B-A17B
86.7Version: TAU2-Bench, from official model card (accuracy)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
8
Command A+
85.0Version: τ²-Bench Telecom (accuracy)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
9
Claude Opus 4.6
84.8Version: τ-benchHarness: Not recordedEvaluator: Not recordedObserved: May 13, 2026Confidence: Not recordedSource
Notes: DAT-5460 source reconciliation: benchlm.ai tau-bench snapshot lists Claude Opus 4.6 at 84.8.
10
GLM-5
82.1Version: τ-benchHarness: Not recordedEvaluator: Not recordedObserved: Apr 24, 2026Confidence: Not recordedSource
11
Qwen3.5-35B-A3B
81.2Version: TAU2-Bench, from official model card (accuracy)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
12
Qwen3.5-122B-A10B
79.5Version: TAU2-Bench, from official model card (accuracy)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
13
Qwen3.5-9B
79.1Version: TAU2-Bench, from official model card (accuracy)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
14
Qwen3.5-27B
79.0Version: TAU2-Bench, from official model card (accuracy)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
15
Grok 4
78.9Version: τ-benchHarness: Not recordedEvaluator: Not recordedObserved: Apr 24, 2026Confidence: Not recordedSource
16
Grok 4.20
78.9Version: τ-benchHarness: Not recordedEvaluator: Not recordedObserved: May 13, 2026Confidence: Not recordedSource
Notes: DAT-5460 source reconciliation: benchlm.ai tau-bench snapshot lists Grok 4.20 at 78.9.
17
GPT-5.4
78.3Version: τ-benchHarness: Not recordedEvaluator: Not recordedObserved: Apr 24, 2026Confidence: Not recordedSource
18
GPT-5.3-Codex
77.8Version: τ-benchHarness: Not recordedEvaluator: Not recordedObserved: Apr 24, 2026Confidence: Not recordedSource
19
Qwen3-Max
76.8Version: τ-benchHarness: Not recordedEvaluator: Not recordedObserved: Apr 24, 2026Confidence: Not recordedSource
20
Qwen3.6-Plus
76.8Version: TAU-bench rank 16 of 37 (pass_rate)Harness: Not recordedEvaluator: Not recordedObserved: Jun 7, 2026Confidence: Not recordedSource
21
Gemini 3.1 Pro Preview
76.5Version: τ-benchHarness: Not recordedEvaluator: Not recordedObserved: Apr 24, 2026Confidence: Not recordedSource
22
GPT-5.2
75.1Version: τ-benchHarness: Not recordedEvaluator: Not recordedObserved: Apr 24, 2026Confidence: Not recordedSource
23
Kimi K2.5
74.2Version: τ-benchHarness: Not recordedEvaluator: Not recordedObserved: Apr 24, 2026Confidence: Not recordedSource
24
Gemini 3 Flash
71.5Version: τ-benchHarness: Not recordedEvaluator: Not recordedObserved: Apr 24, 2026Confidence: Not recordedSource
25
Mistral Large 3 675B Instruct
70.2Version: τ-benchHarness: Not recordedEvaluator: Not recordedObserved: Apr 24, 2026Confidence: Not recordedSource
Showing the top 25 of 29 tracked models. Browse all models.
How to read this benchmark
This benchmark scores models where higher is better. Use scores for directional filtering and shortlisting, not universal quality ranking; then validate pricing, context window, provider availability, and fit for your workload.
Trust this score when
- There is a fresh timestamped snapshot (or multiple snapshots) for this benchmark.
- The model list covers the same version family you can actually deploy today.
- Top candidates overlap with your required routing and feature requirements.
Be cautious when
- There is only one benchmark snapshot or the dataset appears stale.
- The benchmark metric direction is opposite of your decision objective.
- The score difference between options is narrow and likely within implementation variance.
Related benchmarks
Last reviewed: Apr 26, 2026