The Best LLMs for Tool Use on tau2-bench
Which model is most reliable at calling tools?
72 of the 128 buyable models publish a tau2-bench score. It measures multi-turn function calling against a simulated user, which is a different question from writing code.
Method: Published tau2-bench score, descending, exactly as reported. tau2-bench scores multi-turn tool use against a simulated user, so it measures whether a model calls the right function with the right arguments, not whether it writes good code. Pricing as of July 2026. 22 rows below carry a score measured at a non-default reasoning tier and are marked as such; every affected model is named on the hub. Read the full method.
| Rank | Model | tau2-bench | Terminal-Bench | Blended $/1M | Context |
|---|---|---|---|---|---|
| #1 | GLM-5.2Zhipu· non-default tier | 99.1 | 77.9 | $2.15 | 1M |
| #2 | GLM-4.7-FlashXZhipu· non-default tier | 98.8 | not published | $0.152 | 200K |
| #3 | Step 3.7 FlashStepFun | 98.5 | 39.3 | $0.438 | 262K |
| #4 | GLM-5 TurboZhipu | 98.5 | not published | $1.90 | 200K |
| #5 | Claude Fable 5Anthropic | 98.5 | 84.6 | $20.00 | 1M |
| #6 | GLM-5Zhipu· non-default tier | 98.2 | not published | $1.55 | 200K |
| #7 | Qwen3.6-PlusAlibaba | 97.7 | 61.4 | $1.13 | 1M |
| #8 | GLM-5.1Zhipu· non-default tier | 97.7 | 61.8 | $2.15 | 200K |
| #9 | GLM-4.7Zhipu· non-default tier | 95.9 | 45.3 | $1.00 | 200K |
| #10 | Kimi K2.5Moonshot· non-default tier | 95.9 | 45.7 | $1.20 | 262K |
| #11 | Kimi K2.6Moonshot | 95.9 | 65.9 | $1.71 | 262K |
| #12 | Qwen3.6-Max-PreviewAlibaba | 95.9 | not published | $2.92 | 262K |
| #13 | DeepSeek V4-FlashDeepSeek· non-default tier | 95.6 | 56.9 | $0.330 | 1M |
| #14 | Gemini 3.5 FlashGoogle | 95.6 | 76.2 | $3.38 | 1.05M |
| #15 | MiniMax M2.5MiniMax | 95.3 | not published | $0.525 | 128K |
| #16 | MiMo-V2-ProXiaomi | 95.0 | not published | $1.50 | 1.05M |
| #17 | Qwen3.7 MaxAlibaba | 94.7 | 74.5 | $3.75 | 1M |
| #18 | Claude Opus 4.8Anthropic | 94.4 | 84.6 | $10.00 | 1M |
| #19 | MiMo-V2.5-ProXiaomi | 94.2 | 65.2 | $0.544 | 1.05M |
| #20 | DeepSeek V4-ProDeepSeek· non-default tier | 94.2 | 64.8 | $0.990 | 1M |
| #21 | Mistral Medium 3.5Mistral | 94.2 | 50.6 | $3.00 | 256K |
| #22 | Qwen3.5-27BAlibaba· non-default tier | 93.9 | not published | $0.825 | 262K |
| #23 | Qwen3.7 PlusAlibaba | 93.0 | 61.0 | $0.700 | 1M |
| #24 | Grok 4.20xAI· non-default tier | 93.0 | not published | $1.56 | 1M |
| #25 | Claude Opus 4.6Anthropic· non-default tier | 92.1 | not published | $10.00 | 1M |
| #26 | GPT-5.5OpenAI | 91.8 | 80.5 | $11.25 | 1.05M |
| #27 | Grok 4.3xAI | 91.2 | not published | $1.56 | 1M |
| #28 | Kimi K2.7 CodeMoonshot | 90.1 | 67.4 | $1.71 | 262K |
| #29 | Claude Opus 4.5Anthropic· non-default tier | 89.5 | not published | $10.00 | 200K |
| #30 | MiniMax M3MiniMax | 88.9 | 65.2 | $1.05 | 1M |
| #31 | Claude Opus 4.7Anthropic· non-default tier | 88.6 | 83.1 | $10.00 | 1M |
| #32 | Qwen3.5-Omni PlusAlibaba | 88.3 | not published | $1.50 | 262K |
| #33 | Qwen3.5-9BAlibaba· non-default tier | 86.8 | 29.2 | $0.113 | 262K |
| #34 | GPT-5OpenAI | 86.5 | not published | $3.44 | 400K |
| #35 | GPT-5.3 CodexOpenAI | 86.0 | not published | $4.81 | 400K |
| #36 | MiniMax M2.7MiniMax | 84.8 | 55.4 | $0.525 | 205K |
| #37 | Qwen3.5-Omni FlashAlibaba | 84.5 | not published | $0.850 | 262K |
| #38 | GPT-5.1OpenAI· non-default tier | 81.9 | 52.4 | $3.44 | 400K |
| #39 | Nex-N2-ProNex AGI | 81.6 | 67.8 | $1.00 | 262K |
| #40 | GPT-5.6 SolOpenAI | 81.0 | 86.1 | $8.00 | 1.05M |
| #41 | o3OpenAI | 80.7 | not published | $3.50 | 200K |
| #42 | Command ACohere | 80.7 | 22.8 | $4.38 | 128K |
| #43 | Command A+Cohere | 80.7 | 22.8 | $4.38 | 128K |
| #44 | Qwen3 Coder NextAlibaba | 79.5 | 38.2 | $0.290 | 262K |
| #45 | Nova 2.0 LiteAmazon | 75.7 | not published | $0.850 | 1M |
| #46 | Claude Sonnet 4.6Anthropic· non-default tier | 75.7 | 71.2 | $6.00 | 1M |
| #47 | GPT-5.4OpenAI· non-default tier | 74.6 | not published | $5.63 | 1.05M |
| #48 | GPT-5.2OpenAI | 74.3 | not published | $4.81 | 400K |
| #49 | GPT-5.6 TerraOpenAI | 72.8 | 72.3 | $4.50 | 1.05M |
| #50 | GPT-5 MiniOpenAI | 71.1 | not published | $0.688 | 400K |
| #51 | Mercury 2Inception Labs | 70.8 | 27.3 | $0.375 | 128K |
| #52 | o1OpenAI | 62.6 | not published | $26.25 | 200K |
| #53 | o4 MiniOpenAI | 55.6 | not published | $1.93 | 200K |
| #54 | Gemini 2.5 ProGoogle | 54.1 | 28.5 | $3.44 | 1.05M |
| #55 | GPT-4.1 MiniOpenAI | 52.9 | 10.1 | $0.700 | 1.05M |
| #56 | GPT-5.4 NanoOpenAI | 52.6 | not published | $0.463 | 400K |
| #57 | GPT-4.1OpenAI | 47.1 | not published | $3.50 | 1.05M |
| #58 | Mistral Small 4Mistral· non-default tier | 41.2 | 21.0 | $0.263 | 256K |
| #59 | GPT-5.4 MiniOpenAI | 36.5 | not published | $1.69 | 400K |
| #60 | Gemma 4 12BGoogle· non-default tier | 36.3 | 27.3 | $0.00 | 262K |
| #61 | Gemini 2.5 FlashGoogle· non-default tier | 31.6 | not published | $0.850 | 1.05M |
| #62 | Gemini 3.1 Flash-LiteGoogle | 31.3 | 31.1 | $0.563 | 1.05M |
| #63 | GPT-5 NanoOpenAI | 30.4 | not published | $0.138 | 400K |
| #64 | o3 MiniOpenAI | 28.7 | not published | $1.93 | 200K |
| #65 | Ministral 3 14BMistral | 27.2 | 9.7 | $0.200 | 262K |
| #66 | Ministral 3 8BMistral | 26.6 | 4.1 | $0.150 | 262K |
| #67 | GPT-4oOpenAI· non-default tier | 25.1 | not published | $4.38 | 128K |
| #68 | Mistral Large 3Mistral | 24.6 | 12.0 | $0.750 | 262K |
| #69 | Gemini 2.5 Flash-LiteGoogle· non-default tier | 18.4 | not published | $0.175 | 1.05M |
| #70 | Llama 4 MaverickMeta | 17.8 | 7.9 | $0.415 | 1.05M |
| #71 | GPT-4.1 NanoOpenAI | 17.3 | 3.7 | $0.175 | 1.05M |
| #72 | Llama 4 ScoutMeta | 15.5 | 3.7 | $0.135 | 1.05M |