Microsoft AI MAI-Thinking-1 46.0% microsoft.ai for MAI-Thinking-1 on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported August 12, 2026 Technical report Microsoft AI Self-reported Microsoft Record for MAI-Thinking-1 on Terminal-Bench 2.0 70003 Cursor Composer 2.5 69.3% cursor.com for Composer 2.5 on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported May 18, 2026 Model page Cursor Self-reported Record for Composer 2.5 on Terminal-Bench 2.0 50004 DeepSeek AI DeepSeek-V4-Flash (high)56.6% huggingface.co for DeepSeek-V4-Flash (high) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Flash on Terminal-Bench 2.0 110001 DeepSeek AI DeepSeek-V4-Flash (non-thinking)49.1% huggingface.co for DeepSeek-V4-Flash (non-thinking) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Flash on Terminal-Bench 2.0 110001 DeepSeek AI DeepSeek-V4-Flash (max)56.9% huggingface.co for DeepSeek-V4-Flash (max) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Flash on Terminal-Bench 2.0 110001 DeepSeek AI DeepSeek-V4-Pro (high)63.3% huggingface.co for DeepSeek-V4-Pro (high) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on Terminal-Bench 2.0 110002 DeepSeek AI DeepSeek-V4-Pro (non-thinking)59.1% huggingface.co for DeepSeek-V4-Pro (non-thinking) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on Terminal-Bench 2.0 110002 DeepSeek AI DeepSeek-V4-Pro (max)67.9% huggingface.co for DeepSeek-V4-Pro (max) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported April 24, 2026 Model card DeepSeek Self-reported DeepSeek AI Record for DeepSeek-V4-Pro on Terminal-Bench 2.0 110002 OpenAI GPT-5.5 (xhigh)82.7% openai.com for GPT-5.5 (xhigh) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported April 23, 2026 Launch post OpenAI Self-reported Record for GPT-5.5 on Terminal-Bench 2.0 10008 Moonshot AI Kimi K2.6 (thinking)66.7% huggingface.co for Kimi K2.6 (thinking) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported April 20, 2026 Model card Moonshot AI Self-reported Record for Kimi K2.6 on Terminal-Bench 2.0 120002 Anthropic Claude Opus 4.7 (no thinking)69.4% www-cdn.anthropic.com for Claude Opus 4.7 (no thinking) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported April 16, 2026 System card Anthropic Self-reported Record for Claude Opus 4.7 on Terminal-Bench 2.0 20010 Meta AI (originally Facebook AI Research) Muse Spark (thinking)59.0% ai.meta.com for Muse Spark (thinking) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported April 8, 2026 Evaluation methodology Meta Self-reported Record for Muse Spark on Terminal-Bench 2.0 80001 Cursor Composer 1 40.0% cursor.com for Composer 1 on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported March 27, 2026 Technical report Cursor Self-reported Record for Composer 1 on Terminal-Bench 2.0 50001 Cursor Composer 1.5 47.9% cursor.com for Composer 1.5 on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported March 27, 2026 Technical report Cursor Self-reported Record for Composer 1.5 on Terminal-Bench 2.0 50002 Cursor Composer 2 61.7% cursor.com for Composer 2 on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported March 27, 2026 Technical report Cursor Self-reported Record for Composer 2 on Terminal-Bench 2.0 50003 MiniMax MiniMax M2.7 57.0% minimax.io for MiniMax M2.7 on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported March 18, 2026 Launch post MiniMax Self-reported Record for MiniMax M2.7 on Terminal-Bench 2.0 140002 NVIDIA NVIDIA Nemotron 3 Super 120B-A12B 31.0% huggingface.co for NVIDIA Nemotron 3 Super 120B-A12B on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported March 11, 2026 Model card NVIDIA Self-reported Record for NVIDIA Nemotron 3 Super 120B-A12B on Terminal-Bench 2.0 60002 OpenAI GPT-5.4 (xhigh)75.1% openai.com for GPT-5.4 (xhigh) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported March 5, 2026 Launch post OpenAI Self-reported Record for GPT-5.4 on Terminal-Bench 2.0 10007 Google (DeepMind) Gemini 3.1 Pro (High)68.5% deepmind.google for Gemini 3.1 Pro (High) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported February 19, 2026 Model card Google DeepMind Self-reported Record for Gemini 3.1 Pro on Terminal-Bench 2.0 30005 Anthropic Claude Sonnet 4.6 (no thinking, max effort)59.1% www-cdn.anthropic.com for Claude Sonnet 4.6 (no thinking, max effort) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported February 17, 2026 System card Anthropic Self-reported Record for Claude Sonnet 4.6 on Terminal-Bench 2.0 20009 OpenAI GPT-5.3-Codex (xhigh)77.3% openai.com for GPT-5.3-Codex (xhigh) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported February 5, 2026 Launch post OpenAI Self-reported Record for GPT-5.3-Codex on Terminal-Bench 2.0 10006 Anthropic Claude Opus 4.6 (adaptive thinking, max)65.4% www-cdn.anthropic.com for Claude Opus 4.6 (adaptive thinking, max) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported February 5, 2026 System card Anthropic Self-reported Record for Claude Opus 4.6 on Terminal-Bench 2.0 20008 Moonshot AI Kimi K2.5 (non-thinking)50.8% huggingface.co for Kimi K2.5 (non-thinking) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported January 27, 2026 Model card Moonshot AI Self-reported Record for Kimi K2.5 on Terminal-Bench 2.0 120001 Z.ai GLM-4.7 41.0% z.ai for GLM-4.7 on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported December 22, 2025 Launch post Z.ai Self-reported Record for GLM-4.7 on Terminal-Bench 2.0 150001 Mistral AI Devstral 2 32.6% huggingface.co for Devstral 2 on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported December 9, 2025 Model card Mistral AI Self-reported Record for Devstral 2 on Terminal-Bench 2.0 90002 Mistral AI Devstral Small 2 22.5% huggingface.co for Devstral Small 2 on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported December 9, 2025 Model card Mistral AI Self-reported Record for Devstral Small 2 on Terminal-Bench 2.0 90003 Anthropic Claude Opus 4.5 (extended thinking (128K))59.8% assets.anthropic.com for Claude Opus 4.5 (extended thinking (128K)) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported November 24, 2025 System card Anthropic Self-reported Record for Claude Opus 4.5 on Terminal-Bench 2.0 20007 Google (DeepMind) Gemini 3 Pro (high)54.2% blog.google for Gemini 3 Pro (high) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported November 18, 2025 Launch post Google Self-reported Google DeepMind Record for Gemini 3 Pro on Terminal-Bench 2.0 30008 Anthropic Claude Haiku 4.5 (extended thinking (32K))41.8% anthropic.com for Claude Haiku 4.5 (extended thinking (32K)) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported October 15, 2025 Launch post Anthropic Self-reported Record for Claude Haiku 4.5 on Terminal-Bench 2.0 20006 Anthropic Claude Haiku 4.5 (no thinking)40.2% anthropic.com for Claude Haiku 4.5 (no thinking) on Terminal-Bench 2.0 (opens in a new tab) Terminal-Bench Accuracy Reported October 15, 2025 Launch post Anthropic Self-reported Record for Claude Haiku 4.5 on Terminal-Bench 2.0 20006