Benchmark
TAU3-Bench
Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages. Higher is better.
- Models measured
- 10
- Scores recorded
- 14
- Sources
- 2
- IQ weight
- 1.12
Score distribution
- 64.9%
- Mean
- 70.6%
- Median
- 14.6%
- Std dev
- 45.8%
- 10th percentile
- 72.9%
- 90th percentile
- 22.6%
- Min
This benchmark has a weight of 1.12 for chat models, measured over 9 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 1 | GPT-Live 1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 86.2% |
| 2 | MiMo V2.5 ProXiaomi MiMo-V2.5-Pro official launch table · aggregate/domain and harness unspecified· Reported by Xiaomi MiMo· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 72.9% |
| 3 | MiMo V2.5 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 72.9% |
| 4 | Nemotron 3 UltraNeMo Evaluator SDK · TauBench v3 · four-domain average· Reported by NVIDIA· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 70.9% |
| 5 | Qwen3.6 PlusQwen official release table · thinking mode unless benchmark requires otherwise · official user model GPT-5.2 low · default BM25 retrieval· Reported by Alibaba Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 70.7% |
| 6 | Qwen3.6 Plus· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 70.7% |
| 7 | GLM 5.1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 70.6% |
| 8 | GLM 5.1GLM-5.1 official release table· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 70.6% |
| 9 | Qwen3.6 35B A3B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 67.2% |
| 10 | Qwen3.6 FlashQwen3.6-35B-A3B official model card · hosted API name Qwen3.6-Flash· Reported by Alibaba Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 67.2% |
| 11 | Granite 4.2 30B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 62% |
| 12 | Granite 4.2 8B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 58.1% |
| 13 | Granite 4.2 3B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 45.8% |
| 14 | Nemotron 3 Ultra· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 22.6% |
Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · Data by LLM Stats