Product signal
OpenAI publicly names next major model Astra
OpenAI has publicly named Astra as its next major model and used ten results on long-open mathematics and theoretical-computer-science problems as an initial capability record. Astra is not yet publicly deployed.
OpenAI has publicly named its next major model, Astra, and used ten results on long-open mathematics and theoretical-computer-science problems as a capability demonstration. The higher-order change is not simply that an AI system solved more problems. OpenAI is beginning to build a public capability record for its next-generation model through inspectable papers and Lean certificates. Astra is not yet available, and OpenAI has not called it GPT-6.
Astra moves from an internal model to a public capability signal
OpenAI's official page directly describes Astra as “our next major model.” In May, the company disclosed a single unit-distance result from an unnamed internal system. This release ties the named model to ten mathematics and theoretical-computer-science results and estimates the problem-solving tokens at roughly $2,000 using Sol API rates. The important state change is that an officially named next-generation model now has a public portfolio of capability evidence rather than one isolated research result.
Ten results turn the claim into inspectable objects
The 249-page collection covers sphere packing, coding, group theory, complexity, quantum computing, lattice cryptography, and extremal combinatorics, with concrete claims involving a sphere-packing exponent, permanent lower bounds, approximation hardness, and Ramsey-number growth. OpenAI's Selected Proofs repository also provides Lean files, an All.lean file, and Comparator challenge configurations. These artifacts do not replace peer review, but they are closer than an anonymous benchmark to evidence that third parties can run, inspect, and reuse result by result.
A public name is not a deployment announcement
OpenAI has not disclosed Astra's parameters, architecture, API, pricing, or release date. “Next major model” therefore does not establish that Astra is GPT-6 or indicate when it might enter ChatGPT or the API. The Leiden Declaration also notes that automated arguments and translation between natural-language and formal statements can fail. The most accurate description today is an officially named next-generation model with its first public research evidence, but no public deployment.
What to watch next
Two tracks now matter: whether OpenAI publishes an Astra system card, API, pricing, or product access; and whether independent mathematicians compile the Lean files, validate the ten results, and carry them into preprints, seminars, or peer-reviewed literature. Independent replication combined with public deployment would turn Astra from a research preview into a genuine next-generation platform signal.
Sources
- OpenAI Blog — Ten advances in mathematics and theoretical computer science
- OpenAI — Ten Advances in Mathematics and Theoretical Computer Science
- OpenAI — Selected Proofs
- OpenAI — An OpenAI model has disproved a central conjecture in discrete geometry
- Leiden Declaration working group — Leiden Declaration on Artificial Intelligence and Mathematics
- OpenAI — Ten advances in mathematics and theoretical computer science
- Simon Willison — Ten advances in mathematics and theoretical computer science
- The Information — Exclusive: OpenAI Previews ‘Astra’ AI Model in DC
- OpenAI — Previewing GPT-5.6 Sol: a next-generation model
- OpenAI — GPT-5.6 Preview System Card
- OpenAI — Our First Proof submissions
- Eliezer Yudkowsky — OpenAI’s Unreleased Model Astra Solves Ten Major Open Mathematics Problems
- Zvi Mowshowitz — OpenAI's Unreleased Model Astra Solves Ten Major Open Mathematics Problems
- OpenAI — Responding to the next frontier of critical cyber capabilities
- The Decoder AI — OpenAI flags its new Astra model as potentially reaching the highest cybersecurity risk level for the first time
- OpenAI — Preparedness Framework
- OpenAI — OpenAI and Hugging Face partner to address security incident during model evaluation
- UK AI Security Institute — Evidence for inference scaling in AI cyber tasks: Increased evaluation budgets reveal higher success rates
- The Verge — OpenAI puts the brakes on a new model because it’s supposedly too powerful
- Axios — Exclusive: OpenAI slows release of Astra model citing cyber capabilities
- Hugging Face — Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident
- TechCrunch AI — OpenAI says it slowed Astra model development over security concerns
- UK AI Security Institute — Incident Report: unsanctioned agent behaviour during cyber testing
- Bloomberg Technology — OpenAI Pauses Some Work on New Astra Model on Cyber Concerns
- OpenAI — Strengthening cyber resilience as AI capabilities advance
- OpenAI — GPT-5.3-Codex System Card
- The Information — OpenAI Pauses Astra Work on Cyber Concerns
- OpenAI — Our updated Preparedness Framework
- Hugging Face — Security incident disclosure — July 2026
- Engadget — OpenAI slows down Astra development due to cybersecurity concerns
- OpenAI — GPT-5.6 System Card
- Australian Signals Directorate — Careful adoption of Agentic AI in cyber defence
- The Verge — OpenAI hit the brakes. Now what?
- Axios — OpenAI to rewrite its safety rules post-Hugging Face
- OpenAI — Expanding Daybreak as the Cyber Defense Window Narrows
- WIRED AI — OpenAI Overhauls Safety Protocols After Its AI Agents Went Rogue
- OpenAI Blog — Pacing model development in an era of cyber-critical capabilities
- UK AI Security Institute — Our evaluation of OpenAI's GPT-5.5 cyber capabilities