HaiAI123

Curated Global AI Tools Directory

Industry update
Flat

Meta researchers propose SWE-sweep, a benchmark where agents find bugs themselves

11/20/2026 — 11/20, 17:19·1 sources (all)·1 reports (all)

Key takeaways

  • Meta researchers, including two creators of SWE-bench, introduced SWE-sweep, a benchmark that hands a coding agent a real repository without…

Background & analysis

On November 20, 2026, the AI section of DEV Community published a report on SWE-sweep, a new benchmark from Meta researchers that includes two of the original creators of SWE-bench. Unlike most coding benchmarks, which hand an agent a bug report and then ask it to deliver a fix, SWE-sweep takes a different approach. It gives a coding agent a real repository and withholds any hint about what is broken, where the defect sits, or what kind of bug it is. The agent is expected to locate the bugs on its own and repair them.

The benchmark's scope is substantial: it draws on 4,068 known bugs spread across 100 repositories and 22 languages. Fixes are not scored by the agent's own claims but are verified through hidden tests. According to the report, the strongest model evaluated managed to fix only 4.7 percent of the bugs, an outcome that points to discovery, rather than repair, as the harder challenge under this setup.

As of the available coverage, no further updates, follow-up results, or additional statements about SWE-sweep have been published, so the benchmark's status rests on the findings and figures reported at its introduction.

AI-generated from 1 reports · updated 2 days ago

updream
24-hour heatHeat index 30 · peak 59 · 23h ago
24 hours agonow

Reports on this story headlines open the original

Nov 20
  1. Researchers at Meta, including two of the creators of SWE-bench, have introduced SWE-sweep, a benchmark that gives a coding agent a real repository with no hint of what is broken or what kind of bug it is. Across 4,068 known bugs in 100 repositories and 22 languages, the best model tested fixed only 4.7%, verified by hidden tests.

    DEV Community · AIAI score 67

Other stories people are talking about

How is heat calculated?About the method

Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.

This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.

Surge
Discussion rising fast
New
First report within 6 hours
Rising
Still gathering discussion

Back to AI industry updates →