Models & pricing

Where prices live

The pricing page is the source of truth — live per-token prices for every model, input (prompt) and output (completion) priced separately, quoted per million tokens.

Programmatically:

• GET /v1/models — models available to your key
• GET /api/pricing — the public pricing feed the pricing page renders

How a call is billed

1. The gateway counts prompt tokens and completion tokens for the model's tokenizer.
2. Cost = prompt_tokens × input_price + completion_tokens × output_price (per the model's rates).
3. That USD cost is deducted directly from your balance (see Core concepts).
4. The deduction, token counts, and model appear in Console → Usage Logs per request.

Some models bill extras (cached input at a discount, reasoning tokens as output, per-image or per-second media rates) — the pricing page lists these where applicable.

Choosing a model

• Rankings — live usage-based popularity on Linkex.
• Analytics — blind-tested popularity and value-for-money comparisons (price vs. context length vs. value score).

Cost control tips

• Give every app/agent its own key, so usage logs attribute spend cleanly.
• For agents, set spend caps — they bound funding, which bounds total damage.
• Prefer smaller models for high-volume, low-difficulty steps; route only hard steps to frontier models.