Benchmarks
Benchmarks

SRE-Bench

Codingunit % · normalized over [0, 95]

SRE-Bench — software reverse-engineering tasks: recover a binary's core logic without access to source. Reported pass@1 unless a row states an attempt budget.

#ModelSourceScoreNormalized
Score distribution
2 tracked results across the normalization window
095
Score vs. parameters
Open-weights models, log-x params
No open-weights models with disclosed parameter counts have a score here yet.