Used only by mixture-of-experts configurations. Describe strengths in a sentence — “long-form reasoning over large documents” routes better than “reasoning, long-context”.
Input
$/ 1M
Output
$/ 1M
Some self-hosted models are compute-only — no per-token price. Leave both at 0.
tokens
Prompt and conversation history together. A study's usable memory is set by the smallest context window among the models in a set-up.
Enabled
Experimenters can pick this model right away.
Holds no keys — API keys stay in the server's settings