How to cut LLM API costs without changing your product
Most teams overspend for one of five reasons. Four of them are free to fix today; the fifth is about which price list you are on.
1. Route by task, not by habit
The biggest lever is model choice, not provider choice. A “flash” tier model answers classification, extraction, summarisation and most chat turns at a fraction of the price of a flagship. Reserve the expensive model for work that actually needs it.
Rule of thumb: if a wrong answer costs you less than the token delta, use the cheap model.
2. Cap output tokens
Output tokens are the expensive half of every request. A runaway max_tokens is the single most common cause of surprise bills. Set it to what the UI can actually display.
3. Stop re-sending the same context
Long system prompts and documents get billed on every call. Put stable content first, keep the changing part short, and use cache pricing where it exists instead of rebuilding the prompt each time.
4. Batch the boring parts
Keyword matching, templated replies and cached answers should never reach the model. Teams routinely cut 40–60% of calls this way with no quality loss.
5. Pay per token, not per seat
Subscriptions bill you for idle capacity. Prepaid per-token billing means a quiet month costs almost nothing.
Current price examples (live)
These are our prices per 1M tokens, pulled from the live catalogue. Official list prices are shown for comparison on the comparison page.
| Model | Input $/1M | Output $/1M |
|---|---|---|
| Loading… | ||