Rankings
Snapshot 22 November 2026 · Scores copied from each benchmark's own leaderboard · How the CTO Index works
Score and price
| # | Model | Provider | Terminal-Bench | Price $/M |
|---|---|---|---|---|
| 1 | Copilot Fusion | GitHub | 71.2% | Not listed |
| 2 | Claude Fable 5.2 | Anthropic | 65.8% | $8 / $40 |
| 3 | Claude Opus 5.5 | Anthropic | 65.8% | $4 / $20 |
| 4 | Gemini 4 Carbon | 64.5% | $2 / $10 | |
| 5 | Claude Sonnet 5.5 | Anthropic | 63.9% | $2 / $10 |
| 6 | GPT-6.2 Astra | OpenAI | 62.1% | $8 / $40 |
| 7 | GPT-6.2 Sol | OpenAI | 61.5% | $2 / $10 |
| 8 | Gemini 4 Neon | 61.2% | $1.50 / $7.50 | |
| 9 | GPT-6.1 Sol | OpenAI | 60.9% | $2 / $10 |
| 10 | GPT-6 Astra | OpenAI | 60.3% | $10 / $50 |
| 11 | Kimi K3.1 Pro | Moonshot AI | 59.7% | $2 / $10 |
| 12 | Claude Fable 5.1 | Anthropic | 57.6% | $10 / $50 |
| 13 | Gemini 4 Hydrogen | 56.4% | $1.50 / $7.50 | |
| 14 | Gemini 4 Argon | 55.5% | $2 / $10 | |
| 15 | Claude Opus 5 | Anthropic | 53.3% | $5 / $25 |
| 16 | GPT-6.1 Terra | OpenAI | 53.3% | $1 / $5 |
| 17 | GPT-6 Sol | OpenAI | 50.3% | $2 / $10 |
| 18 | Muse Wonder | Meta | 47.0% | $4 / $20 |
| 19 | Claude Fable 5 | Anthropic | 44.8% | $10 / $50 |
| 20 | GLM-5.3 | Z.ai | 41.2% | $1.40 / $4.40 |
| 21 | Claude Haiku 5.5 | Anthropic | 40.0% | $0.10 / $0.50 |
| 22 | Grok 4.7 | xAI | 38.8% | $2 / $6 |
| 23 | GPT-5.6 Sol | OpenAI | 38.5% | $4 / $20 |
| 24 | DeepSeek V4.1 Flash | DeepSeek | 38.5% | $0.15 / $0.60 |
| 25 | Hy4 | Tencent | 37.9% | Not listed |
| 26 | GLM-5.3-Flash | Z.ai | 37.0% | $0.02 / $0.07 |
| 27 | Muse Ember | Meta | 30.6% | $0.10 / $0.50 |
| 28 | Mistral Large 4 | Mistral | 27.6% | $0.70 / $2.10 |
| 29 | Qwen3.8-Max | Alibaba | 27.0% | $2 / $6 |
| 30 | MiMo-V2.6-Pro | Xiaomi | 27.0% | $0.43 / $0.87 |
| 31 | Claude Opus 4.8 | Anthropic | 23.6% | $5 / $25 |
| 32 | Kimi K3 | Moonshot AI | 23.0% | $3 / $15 |
| 33 | Grok 4.6 | xAI | 19.7% | $1.60 / $4.80 |
| 34 | Gemini 3.8 Flash | 18.2% | $0.75 / $3.75 | |
| 35 | MiMo-V2.6-Flash | Xiaomi | 16.1% | $0.14 / $0.28 |
| 36 | GPT-6 Luna | OpenAI | 15.2% | $0.10 / $0.50 |
| 37 | Muse Spark 1.3 | Meta | 14.8% | $1.25 / $4.25 |
| 38 | Inkling 2 Small | Thinking Machines | 13.0% | $0.30 / $1 |
| 39 | Gemini 3.7 Flash | 12.4% | $0.75 / $3.75 | |
| 40 | Claude Sonnet 5 | Anthropic | 12.1% | $2 / $10 |
| 41 | Beam 1.0 | Reflection AI | 6.4% | $0.30 / $1 |
Other benchmarks
Frequently asked questions
What does Terminal-Bench 4.1 measure?
Whether a model, driving an agent harness, can finish real multi-step tasks in a Linux terminal: building software, fixing environments, running and debugging code. The score is the share of tasks resolved.
Which model has the highest Terminal-Bench 4.1 score?
Copilot Fusion from GitHub leads Terminal-Bench 4.1 at 71.2%, ahead of Claude Fable 5.2 (65.8%).
Why does Terminal-Bench 4.1 matter to engineering leaders?
This is the closest public proxy for how a coding agent behaves inside your repository and CI. It rewards finishing the job, not writing a plausible-looking snippet.
What are the limitations of Terminal-Bench 4.1?
Scores depend on the agent harness (Claude Code, Codex, Antigravity CLI, mini-SWE-agent and others), so part of the gap between models is the harness. Error bars are roughly ±3 points; treat differences smaller than that as ties.