Sci-Rho - Physics (English): leaderboard

Metric: Average-case accuracy (%) on the English physics templates of Sci-Rho: 606 expert-written executable templates per language (mathematics, physics, chemistry, biology and computer science, many drawn from olympiad problems), each rendered as 10 visually and numerically varied image-grounded instances; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 17 models tracked.

Top models

#ModelScore
1GPT-5.4 (Non-reasoning)94.6
2Gemini 2.5 Pro90.4
3Gemini 2.5 Flash (Non-reasoning)87.4
4Qwen 3.5 27B (Non-reasoning)87
5Qwen 3.5 122B A10B (Non-reasoning)86.7
6Llama 4 Scout Instruct78.8
7InternVL3.5-8B73.8
8Qwen 3 VL 8B Instruct70.3
9Gemma 3 12B (IT)65.4
10Molmo2-8B61.9
11Gemma 3 4B (IT)44.3
12Qwen 3 VL 4B Instruct40.4

Interactive version: theaggregate.ai/benchmark?slug=sci-rho-physics-english · How It Works · Data refreshed daily, snapshot 2026-09-29.