The swe-bench leaderboard can no longer order its top entries. a september 15 paper shows coding…
the swe-bench leaderboard can no longer order its top entries. a september 15 paper shows coding agents have converged so tightly that the gaps at the top are noise, and argues for measuring something else.
the benchmark died of success.
Context
The arXiv preprint 2609.17394 (submitted 15 September 2026, accepted at ADMA 2026) audits 254 SWE-bench submissions across four splits without running models. On Verified the top two each resolve 396 of 500, the top ten share 285 successes and 51 failures leaving 164 instances that distinguish outcomes, exact paired McNemar tests separate none of the 29 adjacent top-thirty pairs at alpha 0.05, while the larger Test split separates 14 of 23. A leader-based rule gives three descriptive tiers (two after Holm correction). An open SWE-bench issue (19 July 2026) shows the same kind of test on the top eight, with ranks 1 and 2 both at 396 of 500.
The paper says non-rejection does not establish equivalence, so noise overstates it: adjacent differences are not statistically separable at that sample size on Verified, and the Test split partly separates them. Within-model scaffold ranges of 29.8 points against an 8.8 point spread of the top thirty are observational and do not identify causal scaffold effects. It proposes comparison-set-specific resolution reporting, provenance and a five-step audit protocol, not a new benchmark. The submission snapshot date was not read, and only the abstract was read.
Related work
- Earlier note on the same paper ↗Same paper and wording.
- Earlier note on the same audit ↗Same paper, broader framing.
Watch next
- The peer-reviewed version and whether SWE-bench adds confidence intervals or tiers.
Sources
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 21 September 2026 at 17:03 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 →