Terminal-Bench — variant "agent:terminus-1-framework" confirms v1 (Terminus-1 harness), distinct from 2.0/2.1
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Claude Sonnet 4.5 | Anthropic | 51 | 3 | 2026-08-23 |
| 2 | MiniMax M2.1 | MiniMax | 47.9 | 1 | 2026-08-23 |
| 3 | Kimi K2 (Reasoning) | Moonshot | 47.1 | 4 | 2026-06-06 |
| 4 | MiniMax M2 | MiniMax | 46.3 | 2 | 2026-08-23 |
| 5 | GPT-5 | OpenAI | 43.8 | 2 | 2026-06-06 |
| 6 | Claude Opus 4.1 | Anthropic | 43.3 | 1 | 2026-08-23 |
| 7 | Claude 4 Opus | Anthropic | 43.2 | 5 | 2026-08-23 |
| 8 | Claude Haiku 4.5 | Anthropic | 41 | 1 | 2026-08-23 |
| 9 | GLM-4.6 | Z.ai | 40.5 | 2 | 2026-08-23 |
| 10 | GLM-4.5 | Z.ai | 39.9 | 2 | 2026-08-23 |
| 11 | DeepSeek-V3.2 | DeepSeek | 37.7 | 2 | 2026-06-06 |
| 12 | DeepSeek-V3.2-Exp | DeepSeek | 37.7 | 1 | 2026-08-23 |
| 13 | Qwen3 Coder 480B A35B Instruct | Alibaba | 37.5 | 1 | 2026-05-15 |
| 14 | K2-Instruct-0711 | Moonshot | 37.5 | 1 | 2026-05-15 |
| 15 | Claude Sonnet 4 | Anthropic | 36.4 | 6 | 2026-08-23 |
| 16 | Claude 3.7 Sonnet | Anthropic | 35.2 | 1 | 2026-08-23 |
| 17 | LongCat Flash Lite | Meituan | 33.8 | 1 | 2026-08-23 |
| 18 | GLM-4.7 | Z.ai | 33.3 | 1 | 2026-08-23 |
| 19 | DeepSeek-V3.1 | DeepSeek | 31.3 | 2 | 2026-08-23 |
| 20 | MiMo V2 Flash | Xiaomi | 30.5 | 1 | 2026-08-23 |
| 21 | GPT-4.1 | OpenAI | 30.3 | 2 | 2026-06-15 |
| 22 | Kimi K2 Instruct | Moonshot | 30 | 4 | 2026-08-23 |
| 23 | GLM-4.5-Air | Z.ai | 30 | 1 | 2026-08-23 |
| 24 | Nemotron 3 Super 120B A12B | NVIDIA | 25.8 | 1 | 2026-08-23 |
| 25 | Gemini 2.5 Pro | 25.3 | 1 | 2026-06-06 | |
| 26 | NVIDIA Nemotron 3 Nano 30B A3B | NVIDIA | 8.5 | 1 | 2026-08-23 |
| 27 | DeepSeek-R1 | DeepSeek | 5.7 | 1 | 2026-08-23 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.