Visual report of CursorBench 4.0 focused on:
- Claude Opus 5.5
- Claude Fable 5.1
- Grok 4.7
- Muse Spark 1.3
- GPT-6 Astra (estimate)
- GPT-6 Luna (estimate)
Cursor has not published GPT-6 CursorBench scores. Astra and Luna placements are unofficial rough estimates derived from relative performance on FrontierCode 1.1 and Terminal-Bench 4.0 versus models that do have official CursorBench results. Treat them as illustrative, not as Cursor-published numbers.
| File | Description |
|---|---|
| index.html | Self-contained HTML report (charts embedded) |
| cursorbench-4.0-score-vs-cost.png | Score vs cost scatter (log cost) |
| cursorbench-4.0-top-bar.png | Horizontal ranking by score |
| data.json | Plotted points with estimated flags |
| throughput.json | Family average output tokens/s (Artificial Analysis) |
| Model | Score | Cost / task |
|---|---|---|
| Opus 5.5 Max | 57.8% | $13.43 |
| Fable 5.1 Max | 51.8% | $17.28 |
| Grok 4.7 Extra High | 46.3% | $6.01 |
| Muse Spark 1.3 Max | 41.6% | $2.64 |
| Model | Est. score | Est. cost / task |
|---|---|---|
| GPT-6 Astra Max | ~54% | ~$8.50 |
| GPT-6 Luna Max | ~31% | ~$0.22 |
Official data from cursor.com/cursorbench (Sep 2026). Not affiliated with Cursor.