Pricing

Prices are generated from the live catalog — this table is what the platform charges, to the token, with nothing else on the bill.

How pricing works

• Per-token, per-model. Input tokens, output tokens, and cached-input tokens bill at the three rates in the table above — no per-request fee, no minimum, no monthly line item.
• Cache reads bill at the cached rate. Prompt-prefix cache hits are reported in usage.prompt_tokens_details.cached_tokens and billed at the cached column, not the input column.
• Thinking tokens bill as output. On models that emit reasoning (for example the MiniMax family), hidden thinking tokens are billed as output tokens and are visible in usage.completion_tokens_details.reasoning_tokens — the number you are charged for is the number you can audit.
• Vision is available on qwen3.8-27b and muse-glimmer-30b at the same token rates (image parts count as input tokens).

Variant aliases (a -fast tier, for instance) appear on this table the moment they are listed; unlisted variants are never served a price row they were not promised.

What the platform commits to

Three of our beta commitments are price-relevant, stated here in those words:

1. Prepaid wallet = hard cap. Your wallet cannot go negative — runaway billing is structurally impossible.
2. Price you see is price you pay — no credit-load fees. Credits are charged at face value.
3. No mid-term changes to purchased credits. Bought credits keep their terms.

All five commitments live on the beta commitments page.