The top of the coding leaderboard is a tie, not a win. claude opus 5's 97.0% on swe-bench verified…
the top of the coding leaderboard is a tie, not a win. claude opus 5's 97.0% on swe-bench verified sits inside the margin of error with gpt-5.6 sol's 96.2%, and one model family scores 51.9% on one harness and 69.2% on another for the same benchmark.
the leader is whoever ships the better scaffold, not the smarter weights.
Context
Verified source: Vals AI's SWE-bench Verified leaderboard says three of 79 models reach 95% or better, and the leader, Claude Opus 5 at 97.00%, is 3 points from a perfect score, with little room left to separate frontier models. GenzTech's comparison lists Opus 5 at 97.0% and GPT-5.6 Sol at 96.2%.
Modelglass notes that scores are harness-dependent and compares only within the same harness.
The 97.0% and 96.2% match the sources, and Vals AI itself says little room is left to separate frontier models. That the gap is inside the margin of error and the specific 51.9% versus 69.2% harness gap were not seen in the sources read, so unsupported here, not refuted. in the excerpts read, though Modelglass confirms scores depend on the harness. 'The leader is whoever ships the better scaffold' is the author's opinion.
Related work
- Claude Opus 5, GenzTech ↗Rank and score.
- Vals AI leaderboard ↗79 models evaluated.
- Modelglass ↗Harness warning.
Watch next
- Find the harness-by-harness scores for one model to check the 51.9% and 69.2% pair.
Sources
- SWE-bench Verified Leaderboard and Methodology, Vals AIvals.ai
- Claude Opus 5 vs GPT-5.6 Sol, GenzTechgenztech.blog
- Coding Benchmarks, Modelglassmodelglass.com.au
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 30 September 2026 at 23:06 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 →