Benchmark · agentic
CursorBench
Cursor's private, version-tagged internal agent eval built from real Cursor sessions; the trailing number is a version, not a score. Not one of the public suites it is quoted beside.
| Date | Model | Score | Source |
|---|---|---|---|
| 2026-09-01 | Claude Fable 5.1 | 73.4% | Anthropic releases Claude Fable 5.1 with 52.6% on Terminal-Bench-Science and cheaper cache reads |
| 2026-08-13 | Grok 4.6 | 69.9% | SpaceXAI releases Grok 4.6 with 500K context and improved agentic reasoning |
| 2026-07-31 | Claude Opus 5 | 70.0% | Anthropic launches Claude Opus 5; OpenAI models breach Hugging Face during security test |