Coding & terminal-agent ranking
Terminal-Bench 4.0: install the environment, run the tests, read the errors, fix the code — in a real container, with every model on the same scaffold.
| # | Model | Value | Vs. leader | Composite v5 | Output $ | Type |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5Anthropic | 65.2 | #1 | 58.4 | $20.00 | |
| 2 | Gemini 4 CarbonGoogle | 65.2 | 100% | 59.4 | $10.00 | |
| 3 | Claude Sonnet 5.5Anthropic | 64.0 | 98% | 54.2 | $10.00 | |
| 4 | GPT-6.1 TerraOpenAI | 55.6 | 85% | 51.0 | $5.00 | |
| 5 | GLM-5.3 Open weightsZ.ai | 38.9 | 60% | 32.0 | $2.20 | Open weights |
| 6 | Qwen3.8 MaxAlibaba Qwen | 34.3 | 53% | — | $8.00 | |
| 7 | MiMo-V2.6 Pro Open weightsXiaomi | 31.3 | 48% | 30.0 | $1.40 | Open weights |
| 8 | DeepSeek V4.1 Flash Open weightsDeepSeek | 19.7 | 30% | 32.5 | $1.20 | Open weights |
| 9 | Kimi K3 Open weightsMoonshot AI | 17.2 | 26% | — | $15.00 | Open weights |
| 10 | MiniMax M3 Open weightsMiniMax | 1.0 | 2% | — | $1.00 | Open weights |