SciEval (K-12) - Macro-F1: leaderboard

Automatic evaluation of K-12 science lessons on SciEval (2026): given an NGSS-aligned lesson and one EQuIP rubric criterion, the model returns a 0 to 3 score with supporting evidence, compared with expert annotations on the held-out test set. Macro-F1 of the predicted criterion score, in percent; higher is better.

Metric: Macro-F1 (%) of the predicted criterion score (0 not addressed to 3 extensive) against the expert score on the held-out test set of SciEval, a K-12 science instructional-material evaluation benchmark: given an NGSS lesson as page-marked PDF text and one EQuIP rubric criterion, the model returns a 0-3 score and the supporting evidence as JSON, simplified prompt, temperature 0; failed predictions are left out; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 5 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3 4B 2507 Instruct29.5#2221
2Gemini 2.0 Flash26.54#723
3GPT-4o Mini21.7#1667
4Llama 3.1 8B Instruct18.18#4600
5Llama 3.2 3B Instruct17.82#7241

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=scieval-k-12-macro-f1 · How It Works · Data refreshed daily, snapshot 2026-11-21.