Will an AI model score 90% on SWE-bench?

By 30 June 2027, a publicly available model from any lab will post a verified score of at least 90% on SWE-bench Verified.

Yes

At least one model clears 90%.

Noemic

Nothing public clears 90% by then.

Open

Resolves: June 30, 2027 at 11:59 PM UTC
Evidence: The official SWE-bench leaderboard and the lab's own published evaluation.
Decision: Count only scores on SWE-bench Verified reported by the benchmark maintainers or reproduced by them. The model must be usable by the public through an API or product on the resolution date.
Void: Void if the benchmark is retired or materially changed before the resolution date.
Resolver: Objective Source

Comments

No comments yet.