Pick a model for the work

Publisher-reported coding results, one benchmark at a time. Missing results stay missing.

35 routableUpdated

40Models compared
Leaderboard

Coding leaderboard

Choose a benchmark. Each score links to its source.

DeepSWE v1.1

17 models · publisher-reported

Higher scores rank first. Test setups vary by publisher. Prices blend input and output 3:1. Missing results are excluded.

Compare score against price

Scroll to see the full chart →

Method

How we rank

What is ranked
Publisher-reported scores, compared within each benchmark.
What a gap means
No publisher result. MiniRouter never estimates one or counts it as zero.
How to compare
Harnesses, scaffolds and effort settings can differ under one benchmark name.
Next steps

Try it on your code

Use benchmark results to form a shortlist, then check cost and compatibility for the work you actually do.

1. Check the evidence
Open the source for the benchmark you care about. Compare its task, harness and effort settings with your workflow; a higher score does not guarantee a better result on your repository.
2. Estimate the cost
Choose Coding in Model Finder, apply the coding workload example and adjust the token volume. Check tool support, context and unknown rates before choosing a model.
3. Run a small task
Use the OpenCode setup guide to configure your client and choose a supported model. Try a bounded change with tests and review the result and usage before expanding the task.