CursorBench leaderboard: which LLMs score highest
CursorBench (real-world coding-agent tasks) currently has scores for 15 models. The top score is 57.8% by Claude Opus 5.5 (max); the median model scores 41.6% and the lowest scores 27.7%.
| # | Model | Organisation | CursorBench score | Cheapest input $/M | Source | Captured |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 (max) | Anthropic | 57.8% | $4.00 | Epoch AI Benchmarking Hub | 2026-10-06 |
| 2 | Claude Sonnet 5.5 (max) | Anthropic | 55.5% | $2.00 | Epoch AI Benchmarking Hub | 2026-10-06 |
| 3 | Claude Fable 5.1 (max) | Anthropic | 51.8% | $10.00 | Epoch AI Benchmarking Hub | 2026-10-06 |
| 4 | Claude Opus 5 (max) | Anthropic | 46.6% | $5.00 | Epoch AI Benchmarking Hub | 2026-10-06 |
| 5 | Grok 4.7 (xhigh) | xAI | 46.3% | $2.00 | Epoch AI Benchmarking Hub | 2026-10-06 |
| 6 | GLM-5.3 (max) | Z.ai (Zhipu AI) | 42.6% | $0.070 | Epoch AI Benchmarking Hub | 2026-10-06 |
| 7 | GPT-5.6 Sol (max) | OpenAI | 41.7% | $2.00 | Epoch AI Benchmarking Hub | 2026-10-06 |
| 8 | Muse Spark 1.3 (max) | Meta AI | 41.6% | $1.25 | Epoch AI Benchmarking Hub | 2026-10-06 |
| 9 | Grok 4.6 (xhigh) | xAI | 41.4% | $1.25 | Epoch AI Benchmarking Hub | 2026-10-06 |
| 10 | GPT-5.6 Terra (max) | OpenAI | 41.3% | $2.00 | Epoch AI Benchmarking Hub | 2026-10-06 |
| 11 | Gemini 3.8 Flash (high) | Google DeepMind | 39.6% | $0.750 | Epoch AI Benchmarking Hub | 2026-10-06 |
| 12 | GLM-5.3-Flash (max) | Z.ai (Zhipu AI) | 36.8% | $0.036 | Epoch AI Benchmarking Hub | 2026-10-06 |
| 13 | GPT-5.6 Luna (max) | OpenAI | 35.9% | $0.200 | Epoch AI Benchmarking Hub | 2026-10-06 |
| 14 | Claude Sonnet 5 (max) | Anthropic | 34.1% | $2.00 | Epoch AI Benchmarking Hub | 2026-10-06 |
| 15 | Composer 2.5 | Cursor | 27.7% | — | Epoch AI Benchmarking Hub | 2026-10-06 |
Compare all benchmarks side by side on the live leaderboard →
Among the ten highest scorers, the cheapest listed API price belongs to GLM-5.3 (max) at $0.070 per million input tokens (score 42.6%).
What CursorBench measures
CursorBench measures real-world coding-agent tasks.
How to read these scores
Scores are percentages (higher is better): the share of questions or tasks the model completed correctly. Where a model is listed at several reasoning efforts, only its highest-effort row is shown. See the methodology for how rows are chosen.
Where the data comes from
Epoch AI Benchmarking Hub (CC BY 4.0). Scores were last captured 2026-10-06; the Captured column gives each row's date.