Benchmarks · coding
React Native Evals
Coding agents build React Native app features — navigation, animation, async state, lists, native APIs and Skia graphics — and an LLM judge checks the generated code against each task's written requirements. Maintained by Callstack. Score: The board's overall score in percent: the share of requirements met, summed over all 91 evals and every repeat run (an LLM judge decides each requirement against the generated files).
23 models · 23 results · data as of 2026-10-29 · rn-evals.vercel.app ↗
Leaderboard
Opus 5 max
All results
23 rows| # | Model | Effort | React Native Evals |
|---|---|---|---|
| 1 | Opus 5 | max | 90.8% |
| 2 | Fable 5 | max | 90.2% |
| 3 | GPT-5.6 Sol | max | 89.8% |
| 4 | GPT-5.5 | xhigh | 89.3% |
| 5 | Grok 4.5 | max | 88.8% |
| 6 | Grok 4.6 | max | 88.6% |
| 7 | Opus 4.8 | max | 88.5% |
| 8 | GPT-5.6 Terra | max | 87.8% |
| 9 | Sonnet 5 | max | 87.4% |
| 10 | GPT-5.6 Luna | max | 87.4% |
| 11 | Muse Spark 1.1 | max | 86.6% |
| 12 | Gemini 3.1 Pro | high | 84.3% |
| 13 | GLM 5.2 | max | 84.0% |
| 14 | MiniMax M3 | max | 83.5% |
| 15 | Kimi K2.6 | max | 82.9% |
| 16 | Gemini 3.5 Flash | xhigh | 82.3% |
| 17 | Gemma 4 31B IT | high | 80.6% |
| 18 | DeepSeek V4 Flash | max | 80.4% |
| 19 | MiMo V2.5 Pro | xhigh | 80.4% |
| 20 | gpt-oss-120b | high | 75.5% |
| 21 | gpt-oss-20b | high | 73.9% |
| 22 | Mistral Large 3 | none | 72.8% |
| 23 | Nemotron 3 Ultra | max | 67.6% |