OpenAI’s “Astra” reportedly cracked 10 open math and CS problems—let’s wait for the evidence.

Noam Brown posted, HN comments: an internal model may have solved ten major open problems. Nothing is published. We observe, we do not conclude.

In plain terms. OpenAI researcher Noam Brown (@polynoamial) announces that an internal model called Astra has allegedly solved ten major open problems in mathematics and computer science. The proofs have not been published.

The fact

The post, shared on Hacker News (item 49143688) on August 2, 2026, does not list the exact ten problems. One problem (number 6) is said to have been constructed based on the work of researcher Henry Yuen (quantum complexity theory). No written proof is publicly available at the time of writing.

Our take

This announcement is part of a pattern. Recall publication #1342: an OpenAI result on a thirty-year-old open convex optimization problem, announced before peer review. Today’s HN discussion echoes the same criticism—waiting for the proofs to verify the mathematics—and highlights the contrast with standard scientific practices: arXiv deposit, review, replication. Neither alarmism nor hype: we note the announcement, withhold judgment until the proofs are available.

This matters because, if validated, it would shift the bar for what advanced LLMs can actually do in mathematical reasoning—not just mimic derivation chains, but generate verifiable new proofs. This is precisely where the research/hype boundary is at stake.

To watch

Actual publication of proofs (arXiv, OpenAI blog); naming of the ten problems; reactions from Henry Yuen and relevant mathematicians; replication by third parties.