Loading benchmark data…
EyesInAI·Loading live benchmark data
Last tested Jul 1, 2026
Independent indices, shown for context — not our measurement. Artificial Analysis and Design Arena rank models on their own general suites; our pass-rateis task-specific and re-run on the models we actually test. When they line up, that’s corroboration; when they don’t, the task-specific number is usually the one that matters for routing a real workload.
Full precision, max quality. KV cache is only ~10GB even at 1M tokens (sparse attention + Engram memory). Self-host economics close ~3–4B tokens/day — below that the API is cheaper.