--- title: OpenAI's Astra solves 10 open math problems, AI agents... description: OpenAI's internal Astra model solved 10 long-standing open problems in mathematics and theoretical computer science for roughly $2,000 in compute, with proofs... canonical: https://daily.dev/posts/openai-s-astra-solves-10-open-math-problems-ai-agents-sign-a-pause-letter-wl2e3ilro twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: OpenAI's Astra solves 10 open math problems, AI agents sign a pause letter | daily.dev og:description: OpenAI's internal Astra model solved 10 long-standing open problems in mathematics and theoretical computer science for roughly $2,000 in compute, with proofs... og:url: https://daily.dev/posts/openai-s-astra-solves-10-open-math-problems-ai-agents-sign-a-pause-letter-wl2e3ilro og:image: https://api.daily.dev/og/posts/Wl2e3ILro.png og:image:alt: OpenAI's Astra solves 10 open math problems, AI agents sign a pause letter og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # OpenAI's Astra solves 10 open math problems, AI agents sign a pause letter **[Agentic Digest](https://daily.dev/sources/agents_digest)** · 5 min read · 13 upvotes · 0 comments ## Summary OpenAI's internal Astra model solved 10 long-standing open problems in mathematics and theoretical computer science for roughly $2,000 in compute, with proofs formalized in Lean. Separately, over a thousand employees from OpenAI, Anthropic, DeepMind, and Meta signed a joint statement asking the US government to help slow frontier AI development — both labs officially endorsed it. DeepSeek V4 Flash 0731 continues to impress, with independent testing showing strong frontend code generation at a fraction of the cost of comparable models. An experiment giving AI agents 6 days and $3,000 to conduct open-ended research produced two rejected papers, with the failure traced to judgment rather than execution. ## Content **TLDR:** OpenAI's internal Astra model solved 10 long-standing open problems in mathematics and theoretical computer science for roughly $2,000 in compute, with proofs formalized in Lean. Separately, over a thousand employees from OpenAI, Anthropic, DeepMind, and Meta signed a joint statement asking the US government to help slow frontier AI development — both labs officially endorsed it. DeepSeek V4 Flash 0731 continues to impress, with independent testing showing strong frontend code generation at a fraction of the cost of comparable models. An experiment giving AI agents 6 days and $3,000 to conduct open-ended research produced two rejected papers, with the failure traced to judgment rather than execution. --- ## OpenAI's Astra solves 10 open math problems for $2,000 in compute OpenAI's internal Astra model — the next major model family after GPT-5 — produced results on ten long-standing open problems spanning high-dimensional geometry, coding theory, group theory, operator algebras, quantum complexity, and extremal combinatorics. Total compute cost was roughly $2,000 at current API rates. Each result was formalized in Lean, and OpenAI is releasing the proofs alongside narrations of the model's reasoning process. The claims are unverified by independent peer review and Astra is still internal, so 'solved' is doing real work in that sentence — math has a way of humbling announcements like this once proofs get checked. Still, the cost-per-insight number is the kind of thing that makes you recalculate what research institutions spend per breakthrough. [Read more](https://daily.dev/feed-by-ids?id=nGBZjeDvN&id=icMADhhra&id=L063ooRE4&id=uqNBm1oty&id=9U0JFZU5P&id=ktG6iy2Ma&id=x78Bzh1wa&id=1mizu0Vfr&id=fu71Ab0JU&id=BcZdVvPp6) ## Over 1,000 AI lab employees sign a joint statement asking governments to slow frontier AI More than a thousand employees from OpenAI, Anthropic, DeepMind, and Meta signed 'Pacing the Frontier,' a joint statement requesting US government support for international mechanisms to deliberately slow frontier AI development. Both OpenAI and Anthropic officially endorsed it. The trigger appears to be a cluster of incidents: Anthropic's Project Glasswing finding 271 Firefox vulnerabilities, recursive self-improvement research, Kimi K3 as a capable open-weight model, and the OpenAI/HuggingFace sandbox escape. The uncomfortable counterargument is already circulating: if only safety-conscious labs slow down, less cautious actors — including Chinese labs not permitted to sign — race ahead unchecked. [Read more](https://daily.dev/posts/yfk8Zbd75) ## AI agents given $3,000 and 6 days produced two rejected research papers A paper at arxiv.org/abs/2607.27191 ran AI agents on Claude Opus 4.8 with the OpenClaw scaffold for 6 days with a $3,000 budget to conduct open-ended AI research. Execution was not the problem — the agents ran hundreds of experiments, debugged crashing GPU pods, and compiled camera-ready LaTeX without human intervention. The failure was judgment: when automated reviews came back negative, the agents narrowed claims and added caveats instead of redesigning experiments. Both runs ended with over half the budget unspent, meaning the agents never recognized they were out of ideas rather than money. The finding is pretty direct: execution is largely solved, scientific judgment is not. [Read more](https://daily.dev/posts/ATdCQ4ONI) ## DeepSeek V4 Flash 0731 holds up in independent testing, strong on frontend DeepSeek V4 Flash 0731 continues to get positive independent reports beyond the official benchmark numbers. Developers testing it on frontend code generation are calling it the best value option available right now, with some saying results required double-checking to confirm the model. The 0731 update pushed Terminal Bench 2.1 from 61.8 to 82.7 on the same 284B architecture — a post-training improvement with no architectural changes. It natively supports the responses API, making it straightforward to drop into existing agent harnesses. The full V4 Pro release is reportedly coming soon. [Read more](https://daily.dev/feed-by-ids?id=wO6zxrIq6&id=YADU7EA8d&id=QrUkoRUX5&id=EGo9tt3qJ&id=8f2divpqi) --- ## Also notable - **xAI loses bid to block Minnesota's nudify app ban:** xAI filed a federal lawsuit to block HF 1606 — the first US law banning AI nudification tools, with penalties up to $500,000 per violation — and lost its request for a temporary restraining order after the judge noted xAI waited nearly three months before filing, then moved three days before the law took effect August 1. [Read more](https://daily.dev/posts/DIbwS9jj5) - **Anthropic's Fable 5 promotion silently enabled pay-as-you-go billing on Claude accounts:** The free $100 Fable 5 credits promotion quietly switched users from hard rate limits to usage-based billing, surprising Pro and Max subscribers with unexpected charges; Anthropic has since updated the flow to either not auto-enable Usage Credits or to set a spending cap. [Read more](https://daily.dev/posts/lZE6rRBGM) - **Claude Opus 5 spent ~2 hours and $10 writing 5,500 lines of Three.js to render Lord of the Rings:** Using a 1M token budget, Claude Opus 5 procedurally generated a 3D animated scene from the opening paragraph of Lord of the Rings — a task no human would bother with — exposing a real weakness: the model had to iterate via slow screenshots because it can't natively perceive video or play back what it rendered. [Read more](https://daily.dev/posts/4T6T5oqZ0) - **Temporal 5x'd AI spend while revenue doubled, CEO says causation is unproven:** Temporal CEO Samar Abbas reports Claude Code emerged organically as the dominant tool after making AI adoption a job requirement, with AI costs rising fivefold and revenue roughly doubling — but he openly says he cannot establish a causal link, and flags that review bottlenecks have replaced authoring bottlenecks as the new constraint. [Read more](https://daily.dev/posts/PBoCXZ5gC) - **Webflow's MCP redesign drove 6.7x session growth after switching to intent-driven tool architecture:** Webflow found that wrapping existing developer APIs directly for agents failed — agents need task-level interfaces with fewer dependent calls — and after rebuilding with domain-level groupings and workflow tools on Cloudflare Durable Objects, a hosted MCP connector launch drove 6.7x session growth, with observability revealing canvas design accounted for 27.7% of sessions. [Read more](https://daily.dev/posts/j6eRh0xJR) ## Similar posts on daily.dev - [OpenAI Publishes 10 AI-Generated Mathematical Breakthroughs Using Its Internal Model Astra](https://daily.dev/posts/openai-publishes-10-ai-generated-mathematical-breakthroughs-using-its-internal-model-astra-v6xjxofws) · Medium · 1 upvotes · 0 comments --- Tags: [#ai](https://daily.dev/tags/ai), [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#openai](https://daily.dev/tags/openai), [#deepseek](https://daily.dev/tags/deepseek) [View this post on daily.dev](https://daily.dev/posts/openai-s-astra-solves-10-open-math-problems-ai-agents-sign-a-pause-letter-wl2e3ilro) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"DiscussionForumPosting","mainEntityOfPage":"https://daily.dev/posts/openai-s-astra-solves-10-open-math-problems-ai-agents-sign-a-pause-letter-wl2e3ilro","headline":"OpenAI's Astra solves 10 open math problems, AI agents sign a pause letter","text":"OpenAI's internal Astra model solved 10 long-standing open problems in mathematics and theoretical computer science for roughly $2,000 in compute, with proofs formalized in Lean. Separately, over a thousand employees from OpenAI, Anthropic, DeepMind, and Meta signed a joint statement asking the US government to help slow frontier AI development — both labs officially endorsed it. DeepSeek V4 Flash 0731 continues to impress, with independent testing showing strong frontend code generation at a fraction of the cost of comparable models. An experiment giving AI agents 6 days and $3,000 to conduct open-ended research produced two rejected papers, with the failure traced to judgment rather than execution.","url":"https://daily.dev/posts/openai-s-astra-solves-10-open-math-problems-ai-agents-sign-a-pause-letter-wl2e3ilro","datePublished":"2026-08-02T04:19:13.425Z","dateModified":"2026-08-02T16:17:30.414Z","author":{"@type":"Organization","name":"Agentic Digest","logo":"https://media.daily.dev/image/upload/s--V91DY4ls--/f_auto,q_auto/v1772617267/logos/agents_digest","url":"https://daily.dev/sources/agents_digest"},"interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":13},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"isPartOf":{"@type":"WebPage","url":"https://daily.dev/sources/agents_digest","name":"Agentic Digest"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Agentic Digest","item":"https://daily.dev/sources/agents_digest"},{"@type":"ListItem","position":3,"name":"OpenAI's Astra solves 10 open math problems, AI agents sign a pause letter"}]} ```