The benchmark everyone quotes for coding agents was leaking its own answers. a september 10 audit…
the benchmark everyone quotes for coding agents was leaking its own answers. a september 10 audit of swe-bench pro found the verified subset contaminated, and openai's september 16 critique says nearly a third of its questions have issues.
the leaderboard is now the weakest evidence in the room.
Context
The paper SWE-Bench Pro Verified (arXiv 2609.08149, submitted 8 September 2026) finds two defects in SWE-Bench Pro, reward hacking through answer leakage such as Git history and task quality issues, and releases a corrected 731-instance set with anti-hacking controls and task refinement. OpenAI's post on separating signal from noise in coding evaluations, dated 8 July 2026, says about 30 percent of the tasks are broken, from a pipeline that flagged 200 tasks (27.4 percent) and human annotation that found 249 (34.1 percent).
Pro Verified is the authors' corrected set and not a contaminated verified subset, so the note's phrase conflates it with the earlier SWE-bench Verified contamination finding. The paper is dated 8 September and September 10 matches a secondary post only. OpenAI's page is dated 8 July 2026 and September 16 is an aggregator date; nearly a third is OpenAI's estimate. The two audits use different methods and datasets, so the percentages are not merged. The leaderboard is now the weakest evidence in the room is the author's take.
Watch next
- Independent replication of the 731-instance set.
Sources
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 20 September 2026 at 22:33 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 →