BlogLLM Benchmark Scores as Published, Side by Side
SWE-sweep: Results as Published
Every result the catalog holds for SWE-sweep: 5 models from 3 developers, read from 1 page. Each number is as printed by the company named beside it; nothing is ranked or rescored. Models are listed by developer, newest first.
Area Agentic CodingMaintainer Meta Superintelligence Labs, with authors from Harvard University, University of Washington and Stanford UniversityIntroduced 2026Data version 2026103001, as of 2026-10-30
SWE-sweep asks how many bugs a coding agent can find and fix in a large codebase with no hint about the type or location of any bug. Each task is a real repository at a fixed commit where many bugs coexist, each backed by an upstream issue and pull request; the agent explores, decides what is broken and submits one patch. A bug counts as resolved when its hidden tests from the pull request pass and the original test suite still passes; breaking the suite scores the task at zero. The leaderboard lists 100 repositories and 4,100 bugs, and its score is the share of all bugs resolved.
SWE-sweep
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-6 AstraOpenAI, released 2026-09-03 | 9.5% | Google, Gemini 4 Carbonchecked 2026-10-29 | - |
| Claude Opus 5.5Anthropic, released 2026-09-22 | 3.6% | Google, Gemini 4 Carbonchecked 2026-10-29 | - |
| Claude Fable 5.1Anthropic, released 2026-09-01 | 2.9% | Google, Gemini 4 Carbonchecked 2026-10-29 | - |
| Gemini 4 CarbonGoogle, released 2026-10-28 | 9.6% | Google, Gemini 4 Carbonchecked 2026-10-29 | Harness: mini-swe-agent |
| Gemini 4 ArgonGoogle, released 2026-09-30 | 4.3% | Google, Gemini 4 Carbonchecked 2026-10-29 | - |
Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.