OpenAI's Astra Explained: Multi-Agent Model Family That Cracked 10 Open Math Problems

Bottom line: Astra, the next major model family OpenAI unveiled on August 1, coordinates multiple agents over hours or days — and produced new results on ten math problems that had seen no progress for at least a decade. Researcher Noam Brown announced the results; the release date and name (whether it ships as GPT-6) remain undecided. Astra is also set to be the first frontier model to go through the planned U.S. government pre-release review.

What exactly did Astra solve?

An internal version generated new results across these fields — all problems with no progress on the main result for at least ten years, in most cases far longer:

Field	Field	Field
High-dimensional geometry	Coding theory	Arithmetic circuit complexity
Group theory	Operator algebras	Quantum complexity
Lattice cryptography	Extremal combinatorics	—

This is a different claim from a benchmark score. It isn’t about answering known questions more accurately — it’s about producing mathematical results that did not previously exist.

How is it different from current models?

The difference is multi-agent collaboration plus time.

• Today: one question → one answer, in seconds to minutes
• Astra: multiple agents working as a team on a single problem for hours or days

Since most agent tools today fail on endurance rather than intelligence, this direction could widen what agent automation can realistically cover.

When does it ship, and under what name?

• No release date. Whether it launches as GPT-6 or a variant inside the GPT-5 line is undecided
• Classified as a new model class alongside the existing Sol, Terra, and Luna tiers
• Sam Altman demoed it to administration officials and bipartisan senators in Washington on July 29-30
• First through government pre-review — the planned process requiring official sign-off before public release

What it signals for the frontier race

The competitive axis is shifting from “benchmark score” to “how long and how autonomously can it work.” If Opus 5’s ~96% on SWE-bench marked the ceiling on coding accuracy, Astra proposes a different axis entirely: duration and coordination.

FAQ

Q. Can I use Astra now? No. It’s in testing with no announced release date, and it must clear the U.S. government pre-release review first.

Q. Is Astra just GPT-6? Undecided. OpenAI hasn’t determined whether it ships as GPT-6 or as a variant within the GPT-5 line — but it is classified as a new model class separate from Sol, Terra, and Luna.

Q. Why do the math results matter so much? Because the claim is generation, not retrieval. Producing new results on problems stuck for a decade or more is the strongest argument yet that AI is moving from answer tool to research collaborator.

---

Sources: The Decoder, Softonic, Studio Global AI (August 1, 2026); Noam Brown’s announcement