Calibration
Vibe Code Bench → React Native Evals
React Native Evals is estimated from Vibe Code Bench with a Michaelis–Menten curve fitted on 7 models measured on both: y = 0.9333·x / (0.07259 + x), R² = 0.86, cross-validated error 3.0 pp. It is used for 22 estimates.
| Estimated model | Vibe Code Bench | React Native Evals | Source |
|---|---|---|---|
| Claude Haiku 4.5 Thinking | 11.4% | 57.0% | estimated ± 3.0 pp, low confidence |
| Claude Opus 4.5 Thinking | 20.6% | 69.0% | estimated ± 3.0 pp, low confidence |
| Claude Opus 4.6 (Adaptive) | 53.5% | 82.2% | estimated ± 3.0 pp, medium confidence |
| Claude Sonnet 4.5 Thinking | 22.6% | 70.7% | estimated ± 3.0 pp, low confidence |
| DeepSeek V3.2 (Thinking) | 5.1% | 38.5% | estimated ± 3.0 pp, low confidence |
| Gemini 3 Pro | 14.3% | 61.9% | estimated ± 3.0 pp, low confidence |
| Gemini 4 Argon | 91.9% | 86.5% | estimated ± 3.0 pp, low confidence |
| GLM-4.6 | 3.1% | 27.9% | estimated ± 3.0 pp, low confidence |
| GLM-5 (Reasoning) | 23.4% | 71.2% | estimated ± 3.0 pp, low confidence |
| GPT-5.1 | 24.6% | 72.1% | estimated ± 3.0 pp, low confidence |
| GPT-5.1-Codex | 13.1% | 60.1% | estimated ± 3.0 pp, low confidence |
| GPT-5.1-Codex-Max | 22.2% | 70.3% | estimated ± 3.0 pp, low confidence |
| GPT-5.2 | 53.5% | 82.2% | estimated ± 3.0 pp, medium confidence |
| GPT-5 (high) | 20.1% | 68.6% | estimated ± 3.0 pp, low confidence |
| GPT-5 mini | 14.2% | 61.7% | estimated ± 3.0 pp, low confidence |
| Grok 4.1 Fast (Reasoning) | 1.2% | 13.2% | estimated ± 3.0 pp, low confidence |
| Grok 4 Fast (Reasoning) | 0.0% | 0.0% | estimated ± 3.0 pp, low confidence |
| Kimi K2.5 (Reasoning) | 17.5% | 66.0% | estimated ± 3.0 pp, low confidence |
| MiniMax M2.5 | 14.9% | 62.7% | estimated ± 3.0 pp, low confidence |
| Mistral Large 4 | 78.4% | 85.4% | estimated ± 3.0 pp, low confidence |
| Qwen3.5 Plus | 15.7% | 63.9% | estimated ± 3.0 pp, low confidence |
| Qwen3 Max | 3.5% | 30.4% | estimated ± 3.0 pp, low confidence |