Hardware tier · CLOUD

Hosted subscription leaderboard

The same suite the on-prem boards sit, run over a hosted API (e.g. Claude Code's CLI wire). A capability reference line for the hardware tiers — never a hardware comparison: latency and tok/s here measure the service, not silicon you can buy.

Saturation caveat: frontier models exceed this suite's ceiling — 81% of items score perfect for every frontier model tested, and the paired comparison between the top rows is a statistical tie (MDE ≈ 0.05). These rows locate the ceiling; they do not rank frontier models against each other. The suite's discrimination lives on the hardware tiers it was built for.