NewtonBench - Vanilla Agent: leaderboard

Metric: Symbolic Accuracy (%; 324 tasks, mean of four runs). Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.

Top models

#ModelScore
1GPT-575.9
2Gemini 2.5 Pro65.4
3GPT-5 Mini53.1
4Gemini 2.5 Flash50.6
5O4 Mini47.8
6DeepSeek R143.4
7Qwen 3 235B A22B30.9
8QwQ-32B28.5
9DeepSeek V38.9
10GPT-4.15.8
11GPT-4.1 Mini5.1

Interactive version: theaggregate.ai/benchmark?slug=newtonbench-vanilla-agent · How It Works · Data refreshed daily, snapshot 2026-09-24.