ReactBench (39 tasks): Writing React (28 tasks)
Code
October 2026 classification of the unchanged 39-task ReactBench panel: 28 feature or fix tasks with held-out behavioral checks and React Doctor verification, five trials per task. Includes five tasks previously classified as Fix React; kept separate from the historical 23-task writing subset.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 70.0 |
| Claude Fable 5.1 | 62.9 |
| Claude Fable 5 | 59.3 |
| GPT-5.6 Sol | 54.3 |
| Claude Opus 5 | 51.4 |
| GPT-5.6 Terra | 45.7 |
| GLM-5.3 Flash | 44.3 |
| Kimi K3 | 43.6 |
| Sonnet 5 | 43.6 |
| Opus 4.8 | 42.9 |
| DeepSeek V4.1 Flash | 42.9 |
| Grok 4.5 | 42.1 |
| GPT-5.6 Luna | 41.4 |
| Grok 4.6 | 40.7 |
| DeepSeek V4 Pro 0813 | 40.7 |
| Gemini 3.8 Flash | 40.0 |
| Grok 4.7 | 40.0 |
| GLM-5.3 | 38.6 |
| GLM-5.2 | 35.7 |
| Muse Spark 1.1 | 33.6 |
| Gemini 3.1 Pro Preview | 30.7 |
| Kimi K2.7 Code | 30.0 |
| Muse Spark 1.2 | 27.9 |
| DeepSeek-V4-Flash-0731 | 27.9 |
| Gemini 3.5 Flash | 27.1 |
| Composer 2.5 | 23.6 |
| Inkling-Small | 10.0 |