Loading benchmark data…
EyesInAI·Loading live benchmark data
We have no valid benchmark results for this model yet, so there is no pass rate to report. Pricing and specification above come from the provider catalog.
Measured results only — no recommendation.
The practical self-host sweet spot — MIT, 1M context, fits 2× H200 with room for full context. Its 1M KV cache is only ~10GB thanks to sparse attention.
No benchmark results recorded for this model yet.