Calibration

AA-SciCode → React Native Evals

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

React Native Evals is estimated from AA-SciCode with a linear curve fitted on 6 models measured on both: y = 0.4248·x + 0.5543, R² = 0.56, cross-validated error 4.7 pp. It is used for 24 estimates.

Estimated model	AA-SciCode	React Native Evals	Source
A.X K2	41.0%	72.8%	estimated ± 4.7 pp, medium confidence
Claude Haiku 5.5	55.0%	78.8%	estimated ± 4.7 pp, medium confidence
Claude Opus 5.5	66.9%	83.8%	estimated ± 4.7 pp, low confidence
Claude Sonnet 5.5	61.0%	81.3%	estimated ± 4.7 pp, low confidence
DeepSeek V3 0324	39.0%	72.0%	estimated ± 4.7 pp, medium confidence
DeepSeek V4.1 Flash	51.9%	77.5%	estimated ± 4.7 pp, medium confidence
GPT-6.1 Sol	54.2%	78.5%	estimated ± 4.7 pp, medium confidence
GPT-6 Luna	54.6%	78.6%	estimated ± 4.7 pp, medium confidence
GPT-6 Sol	57.6%	79.9%	estimated ± 4.7 pp, medium confidence
Granite 4.2 30B	37.8%	71.5%	estimated ± 4.7 pp, medium confidence
Granite 4.2 3B	25.3%	66.2%	estimated ± 4.7 pp, low confidence
Grok 4.7	57.4%	79.8%	estimated ± 4.7 pp, medium confidence
K-EXAONE 2.0	42.0%	73.3%	estimated ± 4.7 pp, medium confidence
Ling 3.0 Flash VL	44.2%	74.2%	estimated ± 4.7 pp, medium confidence
Ling 3.0 Tiny	24.2%	65.7%	estimated ± 4.7 pp, low confidence
Ling 3.1 Flash	54.1%	78.4%	estimated ± 4.7 pp, medium confidence
Mercury 2.5	38.5%	71.8%	estimated ± 4.7 pp, medium confidence
MiMo-V2.6-Flash	51.3%	77.2%	estimated ± 4.7 pp, medium confidence
MiMo-V2.6-Pro	60.9%	81.3%	estimated ± 4.7 pp, low confidence
MiniCPM5-2B	26.3%	66.6%	estimated ± 4.7 pp, low confidence
North Mini Code	38.8%	71.9%	estimated ± 4.7 pp, medium confidence
Solar Pro 3	25.5%	66.3%	estimated ± 4.7 pp, low confidence
Solar Pro 4	44.6%	74.4%	estimated ± 4.7 pp, medium confidence
Step 5 Preview	58.9%	80.5%	estimated ± 4.7 pp, low confidence