The best coding model on real enterprise code still fails most tasks, and you can't check the…
the best coding model on real enterprise code still fails most tasks, and you can't check the score. specific labs built real-swe from private licensed codebases, where claude fable 5.1 resolves just 38.8% of tasks, and the repos stay closed so nobody outside the company can verify it.
private-code evals are honest in a way leaderboards stopped being.
Context
Specific Labs' Real-SWE benchmark page, dated September 2026, describes tasks from a private production codebase licensed from a real company, scored as pass@1 averaged over eight runs per task with 95 percent confidence intervals, using each model's native harness at high reasoning. Its leaderboard as read lists GPT-6 Astra on Codex CLI at 46.25 percent, Fable 5.1 on Claude Code at 45.00 percent, and Gemini 3.8 Flash on Gemini CLI at 38.75 percent. Sample tasks are available by request access. An aimodelreport.com article of 14 September 2026, seen as a snippet, reports the best frontier agent solving only 38.8 percent.
The current first-party table shows Fable 5.1 at 45.00 percent and 38.75 percent for Gemini 3.8 Flash. The 38.8 percent matches the 14 September secondary report of an earlier snapshot that was not inspected, so this is a different snapshot, not a refutation. Best is bounded to one vendor-run benchmark of model and harness pairs, where GPT-6 Astra now leads. Private repos by design, with sample access on request, and independent verification was not read. Honest in a way leaderboards stopped being is the author's opinion, and the task count was not found.
Related work
- Earlier note on Real-SWE ↗Same benchmark.
- Note on coding leaderboards ↗Related benchmark discussion.
Watch next
- A dated leaderboard history and independent replication.
Sources
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 20 September 2026 at 05:04 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 →