{"slug": "ai-agents-solve-49-of-50-unpublished-physics-problems-in-deepmind-backed-study", "title": "AI agents solve 49 of 50 unpublished physics problems in DeepMind-backed study", "summary": "GPT-6-Astra solved 49 of 50 unpublished, research-level theoretical physics problems with web search enabled, scoring 98% on the TheoPhysEval benchmark, according to a preprint posted to arXiv on October 9, 2026 by 35 researchers from ICTS, Google DeepMind and other institutions. Without web access, GPT-6-Astra solved 46 of 50, while Opus 5.5 scored 46 of 50 on both the realistic and isolated tracks; the realistic GPT-6-Astra run cost $116.28. The study reports that the models' work led to corrections of some human-written reference solutions, though the preprint has not been peer reviewed and the benchmark covers 50 problems across 11 subfields.", "body_md": "Photo: Tara Winstead / Pexels\n\n# AI agents solve 49 of 50 unpublished physics problems in DeepMind-backed study\n\nA new benchmark from physicists at ICTS, Google DeepMind and other institutions finds GPT-6-Astra cracked nearly every research-level problem it was given\n\nAn AI model just worked through a stack of 50 unpublished, research-level theoretical physics problems. It solved 49 of them.\n\nThat result comes from a new preprint called TheoPhysEval, and the model in question is GPT-6-Astra. With web search turned on, it scored 98%, which is the kind of grade that usually gets a graduate student a handshake and a heavier workload.\n\n## What the benchmark actually tested\n\nThe paper is titled “TheoPhysEval: Assessing AI agents at autonomous theoretical physics problem solving.” It landed on arXiv on October 9, 2026.\n\nThe benchmark contains 50 theoretical physics problems that had never been published. They span 11 subfields, including fluid dynamics and quantum field theory.\n\nThe goal was to see what AI agents can do with no human help at all. Nobody nudged the models along or fixed their mistakes midstream.\n\nPhysicists from institutions such as ICTS and [Google](https://cryptobriefing.com/markets/alphabet/) DeepMind led the work. The author list runs to 35 people from a range of research institutions.\n\nThe researchers tested off-the-shelf models, not custom-built research systems. They designed the criteria to approximate the skills an expert theoretical physicist would need.\n\n## The scoreboard\n\nThe study split testing into two tracks. The “realistic” track let models search the web, the way a working researcher would. The “isolated” track cut them off from the internet entirely.\n\n### AI, tech, and the markets they move—in one daily briefing.\n\nDaily. Free. Join 34,000+ readers across crypto, finance, and policy.\n\nOn the realistic track, GPT-6-Astra solved 49 of the 50 problems. The cost of that run came to $116.28.\n\nOn the isolated track, GPT-6-Astra’s score dropped to 46 out of 50. Web access, in other words, was worth three problems.\n\nOpus 5.5 posted 46 out of 50 on both tracks. Giving it the internet didn’t move its score in either direction.\n\nThen there’s the holdout. One problem went unsolved by every model the researchers tested. Fifty questions, and one of them proved stubborn enough to stump the whole field.\n\n## When the student corrects the answer key\n\nPerhaps the most striking finding is not a score at all. According to the study, the models’ work led to corrections of some of the human-written reference solutions.\n\nThe use of unpublished problems matters here too. Problems that have never appeared in print are harder to have memorized from training data, so a correct answer is more likely to reflect actual reasoning than recall.\n\n## Why this matters beyond the lab\n\nThe cost figure deserves attention as well. A $116.28 run on the realistic track is small next to the time a team of human experts would spend working through 50 research-level problems, though the paper’s scope is performance, not a full cost comparison.\n\nThere are caveats worth keeping in view. This is a preprint, which means it has not yet gone through formal peer review. The benchmark is also 50 problems, a meaningful but finite sample of what theoretical physics involves.\n\nAnd solving a well-posed problem is not the same as choosing which problem to solve. The benchmark measures autonomous problem solving, not the creative work of framing new questions, which remains a core part of what theoretical physicists do.\n\n**Disclosure:** This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our\n\n[Editorial Policy](https://cryptobriefing.com/editorial-policy/).", "url": "https://wpnews.pro/news/ai-agents-solve-49-of-50-unpublished-physics-problems-in-deepmind-backed-study", "canonical_source": "https://cryptobriefing.com/ai-agents-solve-49-of-50-physics-problems/", "published_at": "2026-10-25 10:49:39+00:00", "updated_at": "2026-10-25 11:13:21.692009+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-agents", "ai-research", "large-language-models"], "entities": ["GPT-6-Astra", "Google DeepMind", "ICTS", "TheoPhysEval", "Opus 5.5", "arXiv"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/ai-agents-solve-49-of-50-unpublished-physics-problems-in-deepmind-backed-study", "markdown": "https://wpnews.pro/news/ai-agents-solve-49-of-50-unpublished-physics-problems-in-deepmind-backed-study.md", "text": "https://wpnews.pro/news/ai-agents-solve-49-of-50-unpublished-physics-problems-in-deepmind-backed-study.txt", "jsonld": "https://wpnews.pro/news/ai-agents-solve-49-of-50-unpublished-physics-problems-in-deepmind-backed-study.jsonld"}}