Openai says nearly a third of swe-bench pro's questions are flawed. vendors don't fight each other…
openai says nearly a third of swe-bench pro's questions are flawed. vendors don't fight each other anymore, they audit the test set.
next time a repo posts its score, ask who owns the questions.
Context
OpenAI's post, dated 8 July 2026 by search metadata, estimates that about 30% of SWE-bench Pro tasks are broken. An automated pipeline flagged 200 tasks (27.4%) and human annotation identified 249 (34.1%). The process started with 286 flagged tasks, used Codex-based investigator agents plus human review, and each task was reviewed by five engineers. OpenAI calls the figure an estimate and says the earlier SWE-bench Verified lost signal, which led it to recommend Pro.
Nearly a third is the note's paraphrase of about 30%, within a 27.4 to 34.1% range. The post is dated 8 July 2026, 72 days before the note, and no newer OpenAI statement was found. The audited counts are kept as stated; a dataset total is not inferred from rounded fractions. The figures are OpenAI's own, not independent. Vendors audit the test set is the author's opinion, with no evidence beyond this one post. Secondary coverage was a snippet only. The audited task set and its size are not stated in the passages read: dividing 200 by 27.4% gives roughly 730, but that is arithmetic on rounded percentages and says nothing about which set was audited, so it is not used. The 286 are the flagged subset, not the sample size.
Related work
- OpenAI SWE-bench Pro broken (beckmann.ai) ↗Snippet only.
Watch next
- A dataset revision or response from the benchmark's maintainers.
Sources
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 18 September 2026 at 03:42 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →