LLM Stats leaderboard
84 scores · 37 models · first recorded Oct 2, 2026
higher is better · scale 0–1 · record Tau2 Airline
Best reported score
99.3%LongCat-Flash-Thinking-2601 · LLM Stats leaderboard
41 models tested · 2 sources · 9 source documents · latest report Jun 2, 2026
τ-Bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains
Scores stay separate per source — never merged into one cross-source ranking.
84 scores · 37 models · first recorded Oct 2, 2026
higher is better · scale 0–1 · record Tau2 Airline
| Model | Value (%) | Reported | Score type | Recorded |
|---|---|---|---|---|
| MAI-Code-1-Flash | 71.7% | Jun 2, 2026 | unspecified | Oct 2, 2026 |
| Command A+ | 85% | May 20, 2026 | unspecified | Oct 2, 2026 |
| GPT-5.5 | 98% | Apr 23, 2026 | unspecified | Oct 2, 2026 |
| Muse Spark | 91.5% | Apr 8, 2026 | unspecified | Oct 2, 2026 |
| MiMo-V2-Pro | 96.8% | Mar 18, 2026 | unspecified | Oct 2, 2026 |
| GPT-5.4 mini | 93.4% | Mar 17, 2026 | unspecified | Oct 2, 2026 |
| GPT-5.4 nano | 92.5% | Mar 17, 2026 | unspecified | Oct 2, 2026 |
| Nemotron 3 Super (120B A12B) | 56.3% | Mar 11, 2026 | unspecified | Oct 2, 2026 |
| Nemotron 3 Super (120B A12B) | 62.8% | Mar 11, 2026 | unspecified | Oct 2, 2026 |
| Nemotron 3 Super (120B A12B) | 64.4% | Mar 11, 2026 | unspecified | Oct 2, 2026 |
| GPT-5.4 | 98.9% | Mar 5, 2026 | unspecified | Oct 2, 2026 |
| Mercury 2 | 53% | Feb 24, 2026 | unspecified | Oct 2, 2026 |
| Claude Sonnet 4.6 | 91.7% | Feb 17, 2026 | unspecified | Oct 2, 2026 |
| Claude Sonnet 4.6 | 97.9% | Feb 17, 2026 | unspecified | Oct 2, 2026 |
| LongCat-Flash-Lite | 58.0% | Feb 5, 2026 | unspecified | Oct 2, 2026 |
| Claude Opus 4.6 | 91.9% | Feb 5, 2026 | unspecified | Oct 2, 2026 |
| LongCat-Flash-Lite | 73.1% | Feb 5, 2026 | unspecified | Oct 2, 2026 |
| Claude Opus 4.6 | 99.3% | Feb 5, 2026 | unspecified | Oct 2, 2026 |
| LongCat-Flash-Lite | 72.8% | Feb 5, 2026 | unspecified | Oct 2, 2026 |
| LongCat-Flash-Thinking-2601 | 76.5% | Jan 14, 2026 | unspecified | Oct 2, 2026 |
| LongCat-Flash-Thinking-2601 | 88.6% | Jan 14, 2026 | unspecified | Oct 2, 2026 |
| LongCat-Flash-Thinking-2601 | 99.3% | Jan 14, 2026 | unspecified | Oct 2, 2026 |
| MiniMax M2.1 | 87% | Dec 23, 2025 | unspecified | Oct 2, 2026 |
| Nemotron 3 Nano (30B A3B) | 48% | Dec 15, 2025 | unspecified | Oct 2, 2026 |
| Nemotron 3 Nano (30B A3B) | 56.9% | Dec 15, 2025 | unspecified | Oct 2, 2026 |
| Nemotron 3 Nano (30B A3B) | 42.2% | Dec 15, 2025 | unspecified | Oct 2, 2026 |
| GPT-5.2 | 82% | Dec 11, 2025 | unspecified | Oct 2, 2026 |
| GPT-5.2 | 98.7% | Dec 11, 2025 | unspecified | Oct 2, 2026 |
| Nova 2 Omni | 68.8% | Dec 2, 2025 | unspecified | Oct 2, 2026 |
| Nova 2 Pro | 65.2% | Dec 2, 2025 | unspecified | Oct 2, 2026 |
| Nova 2 Lite | 64.8% | Dec 2, 2025 | unspecified | Oct 2, 2026 |
| Nova 2 Omni | 78.3% | Dec 2, 2025 | unspecified | Oct 2, 2026 |
| Nova 2 Pro | 77.7% | Dec 2, 2025 | unspecified | Oct 2, 2026 |
| Nova 2 Lite | 76.5% | Dec 2, 2025 | unspecified | Oct 2, 2026 |
| Nova 2 Pro | 92.7% | Dec 2, 2025 | unspecified | Oct 2, 2026 |
| Nova 2 Omni | 80% | Dec 2, 2025 | unspecified | Oct 2, 2026 |
| Nova 2 Lite | 76% | Dec 2, 2025 | unspecified | Oct 2, 2026 |
| Claude Opus 4.5 | 88.9% | Nov 24, 2025 | unspecified | Oct 2, 2026 |
| Claude Opus 4.5 | 98.2% | Nov 24, 2025 | unspecified | Oct 2, 2026 |
| GPT-5.1 | 67% | Nov 13, 2025 | unspecified | Oct 2, 2026 |
| GPT-5.1 | 77.9% | Nov 13, 2025 | unspecified | Oct 2, 2026 |
| GPT-5.1 | 95.6% | Nov 13, 2025 | unspecified | Oct 2, 2026 |
| GPT-5.1 Thinking | 67% | Nov 12, 2025 | unspecified | Oct 2, 2026 |
| GPT-5.1 Instant | 67% | Nov 12, 2025 | unspecified | Oct 2, 2026 |
| GPT-5.1 Instant | 77.9% | Nov 12, 2025 | unspecified | Oct 2, 2026 |
| GPT-5.1 Thinking | 77.9% | Nov 12, 2025 | unspecified | Oct 2, 2026 |
| GPT-5.1 Thinking | 95.6% | Nov 12, 2025 | unspecified | Oct 2, 2026 |
| GPT-5.1 Instant | 95.6% | Nov 12, 2025 | unspecified | Oct 2, 2026 |
| MiniMax M2 | 87% | Oct 27, 2025 | unspecified | Oct 2, 2026 |
| Claude Haiku 4.5 | 63.6% | Oct 15, 2025 | unspecified | Oct 2, 2026 |
| Claude Haiku 4.5 | 83.2% | Oct 15, 2025 | unspecified | Oct 2, 2026 |
| Claude Haiku 4.5 | 83% | Oct 15, 2025 | unspecified | Oct 2, 2026 |
| LongCat-Flash-Thinking | 67.5% | Sep 22, 2025 | unspecified | Oct 2, 2026 |
| LongCat-Flash-Thinking | 71.5% | Sep 22, 2025 | unspecified | Oct 2, 2026 |
| LongCat-Flash-Thinking | 83.1% | Sep 22, 2025 | unspecified | Oct 2, 2026 |
| Qwen3-Next-80B-A3B-Thinking | 60.5% | Sep 10, 2025 | unspecified | Oct 2, 2026 |
| Qwen3-Next-80B-A3B-Instruct | 45.5% | Sep 10, 2025 | unspecified | Oct 2, 2026 |
| Qwen3-Next-80B-A3B-Thinking | 67.8% | Sep 10, 2025 | unspecified | Oct 2, 2026 |
| Qwen3-Next-80B-A3B-Instruct | 57.3% | Sep 10, 2025 | unspecified | Oct 2, 2026 |
| Qwen3-Next-80B-A3B-Thinking | 43.9% | Sep 10, 2025 | unspecified | Oct 2, 2026 |
| Qwen3-Next-80B-A3B-Instruct | 13.2% | Sep 10, 2025 | unspecified | Oct 2, 2026 |
| Kimi K2-Instruct-0905 | 56.5% | Sep 5, 2025 | unspecified | Oct 2, 2026 |
| Kimi K2-Instruct-0905 | 70.6% | Sep 5, 2025 | unspecified | Oct 2, 2026 |
| Kimi K2-Instruct-0905 | 65.8% | Sep 5, 2025 | unspecified | Oct 2, 2026 |
| LongCat-Flash-Chat | 58.0% | Aug 29, 2025 | unspecified | Oct 2, 2026 |
| LongCat-Flash-Chat | 71.3% | Aug 29, 2025 | unspecified | Oct 2, 2026 |
| LongCat-Flash-Chat | 73.7% | Aug 29, 2025 | unspecified | Oct 2, 2026 |
| GPT-5 | 62.6% | Aug 7, 2025 | unspecified | Oct 2, 2026 |
| GPT-5 | 81.1% | Aug 7, 2025 | unspecified | Oct 2, 2026 |
| GPT-5 | 96.7% | Aug 7, 2025 | unspecified | Oct 2, 2026 |
| Qwen3-235B-A22B-Thinking-2507 | 58.0% | Jul 25, 2025 | unspecified | Oct 2, 2026 |
| Qwen3-235B-A22B-Thinking-2507 | 71.9% | Jul 25, 2025 | unspecified | Oct 2, 2026 |
| Qwen3-235B-A22B-Thinking-2507 | 45.6% | Jul 25, 2025 | unspecified | Oct 2, 2026 |
| Qwen3-235B-A22B-Instruct-2507 | 44% | Jul 22, 2025 | unspecified | Oct 2, 2026 |
| Qwen3-235B-A22B-Instruct-2507 | 71.3% | Jul 22, 2025 | unspecified | Oct 2, 2026 |
| Kimi K2 Instruct | 56.5% | Jul 11, 2025 | unspecified | Oct 2, 2026 |
| Kimi K2 Instruct | 70.6% | Jul 11, 2025 | unspecified | Oct 2, 2026 |
| Kimi K2 Instruct | 65.8% | Jul 11, 2025 | unspecified | Oct 2, 2026 |
| o3 | 64.8% | Apr 16, 2025 | unspecified | Oct 2, 2026 |
| o3 | 80.2% | Apr 16, 2025 | unspecified | Oct 2, 2026 |
| o3 | 58.2% | Apr 16, 2025 | unspecified | Oct 2, 2026 |
| GPT-4o | 45.5% | Aug 6, 2024 | unspecified | Oct 2, 2026 |
| GPT-4o | 63.4% | Aug 6, 2024 | unspecified | Oct 2, 2026 |
| GPT-4o | 23.5% | Aug 6, 2024 | unspecified | Oct 2, 2026 |
5 scores · 4 models · first recorded Oct 2, 2026
higher is better · accuracy · scale 0–100 · record tau2-bench
| Model | Value (%) | Subset / variant | Test conditions | Reported by | Reported | Document |
|---|---|---|---|---|---|---|
| Gemini 3.1 Pro | 99.3% | tau2_bench | Thinking (High), Telecom | Feb 19, 2026 | google_gemini_3_1_pro_model_card | |
| GLM-5 | 89.7% | tau2_bench | Retail and Telecom prompt adjustment; Airline domain fixes per Claude Opus 4.5 system card | Z.ai | Feb 11, 2026 | zai_glm_5_model_card |
| Qwen3.5-397B-A17B | 86.7% | tau2_bench | thinking (temp 0.6, top_p 0.95, top_k 20) | Qwen | Feb 16, 2026 | qwen3_5_model_card |
| Gemma 4 (31B) | 76.9% | tau2_bench | average over 3 | Mar 11, 2026 | google_gemma_4_model_card |
| Model | Value (%) | Reported | Score type | Recorded |
|---|---|---|---|---|
| Gemma 4 (31B) | 76.9% | Mar 11, 2026 | self-reported | Oct 2, 2026 |
| Gemini 3.1 Pro | 90.8% | Feb 19, 2026 | self-reported | Oct 2, 2026 |
| Gemini 3.1 Pro | 99.3% | Feb 19, 2026 | self-reported | Oct 2, 2026 |
| Qwen3.5-397B-A17B | 86.7% | Feb 16, 2026 | self-reported | Oct 2, 2026 |
| GLM-5 | 89.7% | Feb 11, 2026 | self-reported | Oct 2, 2026 |
every document that defines or reports scores
| other | – | Tau2 Airline– |
| other | – | Tau2 Retail– |
| other | – | Tau2 Telecom– |
| other | Gemini 3.1 ProFeb 19, 2026 | |
| other | Gemma 4 (31B)Mar 11, 2026 | |
| other | Qwen | Qwen3.5-397B-A17BFeb 16, 2026 |
| other | Z.ai | GLM-5.1Apr 3, 2026 |
| other | Z.ai | GLM-5Feb 11, 2026 |
| tech report | Z.ai | GLM-5Feb 11, 2026 |
Records are located via the tracking corpus (ktwu01/benchmark-radar live dataset). Every row cites its source document.