Benchmark every open-weight coding model

No riders yet

Which open model actually ships code?

Run a fair, reproducible coding benchmark across every tool-calling open-weight model available through repo.ing Inference. Use a public suite such as SWE-bench Lite plus a small held-out task set written for this run. Report solve rate, median latency and cost per solved task, with the exact prompts, harness settings and seeds so anyone can rerun it. Flag contamination risks and publish every failure.