12 coding benchmarks. Each score is compared with the same model on Claude Code, earlier Pixis runs, and the top public results reported for other models. Solve rate = tasks solved ÷ tasks in the set.
Each row is one benchmark. Green dots are the top public results for other models; hover any dot for the model and source.
Pixis harness score minus the comparison, in percentage points. Right of zero means Pixis coder pro on the Pixis harness scores higher.
Every run on each benchmark, sorted by score.
Pixis runs and the public results, with sources. Click a column to sort.