Physics-Intern Harness Lifts Gemini 3.1 Pro From 17.7 to 31.4, Topping GPT 5.5 Pro

aiai-modelingai-research-evalsai-productsai-agents-coding 2 posts · 2 accounts

Physics-intern, a newly released harness for science problems, is designed to wrap any model and use a dedicated subagent to improve vanilla reasoning models. Its creator said Gemini 3.1 Pro improved from 17.7 to 31.4 with the tool, surpassing GPT 5.5 Pro, while GPT 5.5 Pro itself did not benefit from the harness.

From the sources (2 posts)

@lvwerra

We released physics-intern: a simple harness for science problems! It gets models like Gemini 3.1 Pro to go from 17.7 -> 31.4, thus beating GPT 5.5 Pro. The physics-intern harness can wrap any model and via dedicated subagent boost the

@_lewtun

RT @lvwerra: We released physics-intern: a simple harness for science problems! It gets models like Gemini 3.1 Pro to go from 17.7 -> 31.…

Senators Probe DOJ’s $1.776 Billion Anti-Weaponization Fund in Trump IRS Settlement HYPE Token Jumps 5%, Returns to $11 Billion Market Cap After Bitwise ETF Debut