📊 Benchmark Lab
Run public benchmark slices through Swiftest pipelines — Challenger vs classic Many Minds vs single-model baselines — on identical items, with repeats and bootstrap confidence intervals. Suites: AIME 2026 (30 competition math problems, exact-match) and GPQA-Diamond (198 PhD-level MCQs, choices reshuffled per repeat).
Benchmarks are available to signed-in accounts (runs are billed in credits — 10 credits per item-run)
and free for admins. Sign in to run one.