ReactBench (39 tasks): Writing React
Code
Writing React subset of the revised ReactBench panel; 23 feature or fix tasks with held-out behavioral checks and React Doctor verification, five trials per task.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 61.7 |
| Claude Fable 5.1 | 60.9 |
| Claude Opus 5 | 55.7 |
| GPT-5.6 Sol | 50.4 |
| GPT-5.6 Terra | 50.4 |
| GLM-5.3 Flash | 44.3 |
| GPT-5.6 Luna | 42.6 |
| Opus 4.8 | 40.9 |
| Kimi K3 | 40.9 |
| Grok 4.6 | 39.1 |
| Sonnet 5 | 37.4 |
| Gemini 3.8 Flash | 36.5 |
| GLM-5.3 | 36.5 |
| DeepSeek V4 Pro 0813 | 34.8 |
| Grok 4.5 | 33.0 |
| Gemini 3.1 Pro Preview | 30.4 |
| GLM-5.2 | 30.4 |
| Gemini 3.5 Flash | 27.8 |
| Kimi K2.7 Code | 27.8 |
| DeepSeek-V4-Flash-0731 | 27.0 |
| Muse Spark 1.1 | 26.1 |
| Muse Spark 1.2 | 25.2 |
| Composer 2.5 | 19.1 |
| Inkling-Small | 9.6 |