OpenAI Astra Explained

Astra is OpenAI's next major model family, but it hasn't been released yet. OpenAI first named it in an August 1, 2026 research post on mathematics. Six days later, the company paused parts of the work following a safety evaluation.

For now, Astra has no announced price, context window, rate limits, or model card. OpenAI has published some results and documented the safety pause. On September 2, 2026, it said it was preparing to release Astra with stronger safeguards and limited access.

For more on timing, see the OpenAI Astra release date page. The ChatGPT 6 features page covers the naming question, while the OpenAI Astra benchmarks page looks at the supporting evidence.

What Is OpenAI Astra?

Astra is the next major model family from OpenAI, announced on August 1, 2026 in a research post about mathematics and theoretical computer science. It is not available to the public.

The name sits alongside the GPT-5.6 lines already in production: Sol, Terra, and Luna. OpenAI has not said whether Astra ships under a GPT-6 label, a GPT-5.7 label, or a name of its own.

OpenAI published no specifications alongside the announcement. There is no rate card, no context window, and no model card, so buyers have nothing to size a budget against.

• Announced — August 1, 2026, in an OpenAI research post on mathematics and theoretical computer science.
• Status — unreleased, with no announced release date, so no team can plan a migration around it yet.
• Naming — undecided in public. OpenAI has not confirmed GPT-6, GPT-5.7, or any other label.
• Positioning — a new family alongside GPT-5.6 Sol, Terra, and Luna, aimed at long-running work split across several agents.
• Access at launch — restricted. OpenAI has said the advanced capabilities go to selected early testers rather than to every account.

What Astra Has Already Proven

Astra's one public result is a set of ten solved problems in mathematics and theoretical computer science, published by OpenAI on August 1, 2026. An internal version of the model produced them.

OpenAI put the compute cost for all ten solutions at roughly $2,000 at GPT-5.6 Sol application programming interface (API) rates. That price is why the result travelled well outside mathematics.

The proofs can be checked by machine rather than taken on trust. OpenAI released a 249-page manuscript and Lean 4 proof certificates on GitHub under an Apache 2.0 licence. In Lean, the keyword sorry marks a step nobody has proved. The repository reports a sorry count of zero, so no step in the ten formalised proofs is left unproved.

• An explicit construction of a non-sofic group, a question open since Mikhail Gromov set out soficity in 1999.
• A disproof of Connes's rigidity conjecture on von Neumann algebras.
• A proof of Ehrhart's volume conjecture.
• Three problems from the Erdos catalogue, including problem 183 on multicolour Ramsey numbers.
• Roughly $2,000 of compute for all ten solutions at GPT-5.6 Sol API rates, which made the result a business story rather than only a mathematics one.

Why OpenAI Paused Work on Astra

OpenAI paused parts of its frontier training in August 2026 after an internal evaluation placed Astra at the Critical cybersecurity capability threshold. That is the highest severity level defined in OpenAI's Preparedness Framework, and Astra is the first model OpenAI has put there.

The evaluation was dated August 7, 2026. It followed a July test, reported by Axios, in which an internal model executed 17,600 autonomous intrusion actions against Hugging Face infrastructure.

The response was concrete rather than a statement of intent. OpenAI told Axios on August 18, 2026 that it had stopped two weeks of deployment-focused reinforcement-learning (RL) training and put its largest planned frontier RL run on hold. It also began rewriting the Preparedness Framework, most of which was written in 2023.

• July 2026 — an internal test produced a model that ran 17,600 autonomous intrusion actions against Hugging Face infrastructure.
• August 7, 2026 — a safety evaluation places Astra at the Critical cybersecurity threshold, the highest level the Preparedness Framework defines.
• August 18, 2026 — OpenAI confirms a two-week stop on deployment-focused RL training and a hold on its largest planned frontier RL run (Axios).
• Framework rewrite — the Preparedness Framework itself is being rewritten, because the 2023 text did not anticipate a model reaching this threshold.
• New controls — earlier alignment and security checks, more monitoring during development, and higher safeguards when scaling post-training.
• September 2, 2026 — OpenAI says it is preparing to release Astra with stronger safeguards, and that advanced capabilities go to selected early testers first.

What OpenAI Has Not Published About Astra

OpenAI has published no pricing, no rate limits, no context window, no model card, and no ChatGPT availability date for Astra. Every figure a buyer needs to size a migration is missing.

GPT-5.6 shipped with per-token prices on the day it reached general availability. Astra has none of that, which is the difference between an announcement and a product.

Confirmed timing milestones, and the difference between what OpenAI has said and what prediction markets are guessing, are tracked on the OpenAI Astra release date page.

• No published price per million tokens, so no team can model what Astra would cost to run.
• No published context window, so no team can tell whether its own documents fit in one call.
• No published rate limits, so capacity planning has nothing to work from.
• No model card, which is where OpenAI normally documents evaluations and known limitations.
• No ChatGPT availability, so Astra cannot be tried through the consumer product today.

Is Astra the Same Thing as GPT-6?

OpenAI has not confirmed that Astra will be called GPT-6. Reporting has floated both GPT-6 and GPT-5.7 as possible labels, and OpenAI has committed to neither.

The confusion is understandable, because Astra is the next major model and GPT-6 is the next number in the public naming sequence. Those are two different claims, and only the first one has a source behind it.

The full naming picture, and what a ChatGPT 6 release is expected to include, sits on the ChatGPT 6 features page.

What Astra Changes for Teams Running GPT-5.6 Today

Nothing about Astra should change a production stack this month. At Layer3Labs, we build and run AI systems inside other people's businesses, and the model launches worth reacting to are the ones with a price and an availability date attached.

Governance is the part to act on. A vendor classifying its own model at a Critical cybersecurity threshold changes what a careful review asks about, and you can update that review before Astra ships.

Who should ignore this entirely: any team still getting value from GPT-5.6 Terra or Luna on routine work. Astra is aimed at long-running problems that need several agents. Ticket triage is the opposite job, and a cheaper tier will keep beating it there regardless of what Astra scores.

What would change this answer: a published price, a published context window, and general availability rather than an early-tester list. Until those three land, there is no migration to plan.

• Do now — add a frontier-capability question to your vendor review, so a Critical classification is something your process already handles.
• Do now — write down which of your workloads would genuinely benefit from long-horizon multi-agent work, so you can test Astra against a real case rather than a demo.
• Wait for — published pricing, because tier routing is a cost decision before it is a capability one.
• Wait for — general availability, since an early-tester programme cannot carry a production workload.
• Do not do — re-platform a working GPT-5.6 workflow in anticipation of a model with no release date.