SWE-fficiency: Results as Published

SWE-fficiency asks an agent to make a real repository faster on a real workload without breaking behavior. The suite has 498 tasks from nine Python libraries, including numpy, pandas, and scipy. Each task provides a codebase, a slow workload, and the repository tests that must still pass. The score is the speedup ratio, the agent's runtime speedup divided by the human patch's speedup, combined across tasks with a harmonic mean, and a patch that fails the tests or does not apply is scored as if no edit was made.

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.