Get startedLeaderboardDecode calculatorModelsOrganizationsReportsMatrix runsHardwareBenchmarksMarketplaceRentalsProAPI Docs
Language
Actual Computer — Every computer, one endpoint

Observed performance

Run snapshot

All runs matching the filters below.
Fastest run
Median speed
Memory ceiling
Slowest run
Measured runs
Qwen3.8-27B-EXL3-3.5bpw — 164.6 tok/s on RTX 3090 · 24 GB | LocalMaxxing