# τ³-bench results leaderboard

Agentic benchmark

> Every figure below is reproduced as its upstream source published it: nothing is modelled, estimated, interpolated or converted. `Unavailable` means no source published the value — it is never a zero. Each value carries its evidence state and the date it was observed.

Page: https://modelscale.dev/leaderboards/benchmarks/agentic-tau3-bench

τ³-Bench Tool-Agent-User Evaluation. Every model the catalog carries a published τ³-bench results value for, ranked by that value.

## Benchmark record

| Field | Value |
| --- | --- |
| Category | Agentic |
| Measure | Published domain or average success results |
| Tasks | Corrected customer-service tasks plus knowledge and voice evaluation modes |
| Difficulty | Long-horizon, multimodal, and knowledge-aware tool use |
| Published by | [Official τ³-bench repository and release notes](https://github.com/sierra-research/tau2-bench) |

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.

## τ³-bench results ranking

32 models with a published τ³-bench results value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.

| Rank | Model | Provider | Published domain or average success results | Evidence |
| --- | --- | --- | --- | --- |
| 1 | [Mercury 2.5](https://modelscale.dev/models/mercury-2-5) | Inception | 96.0 | Observed 2026-11-21 · source benchlm:models |
| 2 | [Mistral Medium 3.5 128B](https://modelscale.dev/models/mistral-medium-3-5-128b) | Mistral | 91.4 | Observed 2026-11-21 · source benchlm:models |
| 3 | [MiMo-V2.5-Pro](https://modelscale.dev/models/mimo-v2-5-pro) | Xiaomi | 72.9 | Observed 2026-11-21 · source benchlm:models |
| 4 | [Nemotron 3 Ultra](https://modelscale.dev/models/nemotron-3-ultra-500b) | NVIDIA | 70.9 | Observed 2026-11-21 · source benchlm:models |
| 5 | [Qwen3.6 Plus](https://modelscale.dev/models/qwen3-6-plus) | Alibaba | 70.7 | Observed 2026-11-21 · source benchlm:models |
| 6 | [GLM-5.1](https://modelscale.dev/models/glm-5-1) | Z.AI | 70.6 | Observed 2026-11-21 · source benchlm:models |
| 7 | [Claude Opus 4.5](https://modelscale.dev/models/claude-opus-4-5) | Anthropic | 70.2 | Observed 2026-11-21 · source benchlm:models |
| 8 | [Qwen3.5 397B](https://modelscale.dev/models/qwen3-5-397b) | Alibaba | 68.4 | Observed 2026-11-21 · source benchlm:models |
| 9 | [LFM3-1.7B](https://modelscale.dev/models/lfm3-1-7b) | LiquidAI | 68.0 | Observed 2026-11-21 · source benchlm:models |
| 10 | [MiniMax M2.7](https://modelscale.dev/models/minimax-m2-7) | MiniMax | 67.6 | Observed 2026-11-21 · source benchlm:models |
| 11 | [Qwen3.6-35B-A3B](https://modelscale.dev/models/qwen3-6-35b-a3b) | Alibaba | 67.2 | Observed 2026-11-21 · source benchlm:models |
| 12 | [Pokee-Isaac 28B](https://modelscale.dev/models/pokee-isaac-28b) | Pokee AI | 66.2 | Observed 2026-11-21 · source benchlm:models |
| 13 | [Kimi K2.5](https://modelscale.dev/models/kimi-k2-5) | Moonshot AI | 65.7 | Observed 2026-11-21 · source benchlm:models |
| 14 | [GLM-5](https://modelscale.dev/models/glm-5) | Z.AI | 65.6 | Observed 2026-11-21 · source benchlm:models |
| 15 | [Granite 4.2 30B](https://modelscale.dev/models/granite-4-2-30b) | IBM | 62.0 | Observed 2026-11-21 · source benchlm:models |
| 16 | [Granite 4.2 8B](https://modelscale.dev/models/granite-4-2-8b) | IBM | 58.1 | Observed 2026-11-21 · source benchlm:models |
| 17 | [Ling 3.1 Flash](https://modelscale.dev/models/ling-3-1-flash) | InclusionAI | 47.8 | Observed 2026-11-21 · source benchlm:models |
| 18 | [Granite 4.2 3B](https://modelscale.dev/models/granite-4-2-3b) | IBM | 45.8 | Observed 2026-11-21 · source benchlm:models |
| 19 | [Atria Dawn Preview](https://modelscale.dev/models/atria-dawn-preview) | Shanghai Artificial Intelligence Laboratory | 41.2 | Observed 2026-11-21 · source benchlm:models |
| 20 | [Beam 1.0 501B-A23B FP8](https://modelscale.dev/models/reflection-beam-fp8) | Reflection AI | 36.8 | Observed 2026-11-21 · source benchlm:models |
| 21 | [Beam 1.0 501B-A23B](https://modelscale.dev/models/reflection-beam) | Reflection AI | 36.1 | Observed 2026-11-21 · source benchlm:models |
| 22 | [Beam 1.0 501B-A23B NVFP4](https://modelscale.dev/models/reflection-beam-nvfp4) | Reflection AI | 35.4 | Observed 2026-11-21 · source benchlm:models |
| 23 | [Spark-X2.6-4B](https://modelscale.dev/models/spark-x2-6-4b) | SparkLLM | 34.7 | Observed 2026-11-21 · source benchlm:models |
| 24 | [Ripple-65B Preview](https://modelscale.dev/models/ripple-65b-preview) | Ricursive Intelligence | 31.3 | Observed 2026-11-21 · source benchlm:models |
| 25 | [Mercury 3](https://modelscale.dev/models/mercury-3) | Inception | 30.2 | Observed 2026-11-21 · source benchlm:models |
| 26 | [Nemotron 3.5 Super VL 120B A12B](https://modelscale.dev/models/nemotron-3-5-super-vl-120b-a12b) | NVIDIA | 18.9 | Observed 2026-11-21 · source benchlm:models |
| 27 | [Nemotron 3.5 Super VL 120B A12B NVFP4](https://modelscale.dev/models/nemotron-3-5-super-vl-120b-a12b-nvfp4) | NVIDIA | 17.8 | Observed 2026-11-21 · source benchlm:models |
| 28 | [Nemotron 3.5 Bolt 74B-A9B](https://modelscale.dev/models/nemotron-3-5-bolt-74b-a9b) | NVIDIA | 17.0 | Observed 2026-11-21 · source benchlm:models |
| 28 | [Nemotron 3.5 Bolt 74B-A9B BF16](https://modelscale.dev/models/nemotron-3-5-bolt-74b-a9b-bf16) | NVIDIA | 17.0 | Observed 2026-11-21 · source benchlm:models |
| 30 | [Nemotron 3.5 Lightning 30B A3B NVFP4](https://modelscale.dev/models/nemotron-3-5-lightning-30b-a3b-nvfp4) | NVIDIA | 9.5 | Observed 2026-11-21 · source benchlm:models |
| 31 | [LFM2.5-24B-A2B](https://modelscale.dev/models/lfm2-5-24b-a2b) | LiquidAI | 8.0 | Observed 2026-11-21 · source benchlm:models |
| 32 | [LFM2.5-2.6B](https://modelscale.dev/models/lfm2-5-2-6b) | LiquidAI | 5.7 | Observed 2026-11-21 · source benchlm:models |

Rows are ordered by the value the source published, highest first. The source does not state whether a higher value is the better result, so this document does not either: for a benchmark that measures a rate of failure, read the table from the bottom.
