OpenAI confirmed Astra as its next major model family on Aug. 1, an unreleased system that solved 10 math problems, arriving a week after Anthropic's Claude Opus 5.
Key Points:
- OpenAI confirmed the Astra name on Aug. 1 inside a research post describing ten solved math problems.
- Astra coordinates several agents on one problem for hours or days, and it remains unreleased with no set date.
- Claude Opus 5 shipped Jul. 24 at $5 and $25 per million tokens, with a 1M-token context window.
OpenAI Astra Model Details
OpenAI announced the name inside a research post crediting an internal version of Astra with ten results in mathematics and theoretical computer science, each open for at least a decade. The work spans high-dimensional geometry, coding theory and quantum complexity, and one proof establishes the existence of non-sofic groups.
The pitch centers on duration rather than raw speed, since Astra coordinates several agents that work a single problem together for hours or even days.
Earlier reporting placed it beside the Sol, Terra and Luna families, with no release date and no decision yet on a GPT-6 or GPT-5.7 label. Sam Altman demoed the system to policymakers in Washington, and the family is expected to be the first reviewed under a planned federal approval framework.
Also Read: AI Infrastructure, Payday Lender's Desperate $1B Pivot To Dominate Data Centers
Claude Opus 5 Pricing And Benchmarks
Anthropic launched Claude Opus 5 on Jul. 24 at $5 per million input tokens and $25 per million output tokens, matching the rate charged for Opus 4.8. It topped the Artificial Analysis Intelligence Index at 61 points while costing roughly half as much per task as Fable 5.
The model runs a 1M-token context window and caps output at 128,000 tokens. It scored 30.2% on ARC-AGI-3, about three times the next-best published result, and 89.1% on Terminal-Bench 2.1 at maximum effort. An Anthropic spokesperson advised buyers to run competing models on one bounded task and one long-horizon job before making a commitment.
Astra Expert Reactions
Astra carries no published price, no benchmark table and no outside testing, which leaves buyers weighing a shipped product against a research claim.
Thomas Bloom, a University of Manchester mathematician, called the ten results "big news" on X and rated them above OpenAI's earlier proof.
Bloom also pushed back on claims that the model replaces mathematicians, since people built it and trained it on everything mathematicians have written. Noam Brown, an OpenAI researcher, noted the run cracked no Millennium Prize Problems and said the team did not spend much on each one.
Cognitive scientist Gary Marcus argued the results are real but badly oversold as a solved field.
The announcement follows a May result from the same long-horizon family, which disproved the Erdos unit distance conjecture, an 80-year-old problem in discrete geometry. Fields Medalist Tim Gowers said then he would recommend that proof for the Annals of Mathematics without hesitation.
Read Next: Foldable iPhone Rumors Get Specific: $2,500, No Face ID, No Telephoto