Perplexity models through odnoga
Models that answer from the live web rather than from training data alone. Use them where the answer has to be current, and price them as what they are: a search and a generation in one call.
When freshness is the requirement
Every other model on the gateway answers from what it learned during training, with a cutoff. These do not — which makes them the right choice for questions about prices, availability, news and anything else that changed this week, and the wrong choice for deterministic work where you want the same answer every time.
Budget them separately
A web-grounded answer does more work than a plain completion, and the cost reflects it. If you expose this to end users, give it its own budget and its own per-user cap rather than pooling it with your cheap traffic — odnoga enforces caps before the spend happens, not after the invoice.
Vendor cost is the list price. Your price is that plus your plan margin — the same figures that appear on your invoice.
Every model here, cheapest first
| Model | Vendor | Context | In / 1M | Out / 1M | Cached in | Your price in | |
|---|---|---|---|---|---|---|---|
| Sonar sonar | Perplexity | 127K | $1 | $1 | — | $1.08 | Details |
| Sonar Pro sonar-pro | Perplexity | 200K | $3 | $15 | — | $3.23 | Details |
Your price column is the vendor cost +7%.
Narrow it differently
Models that can decide to call a function you defined and use the result.
The widest range on the gateway, and the widest price range with it — from the cheapest token on odnoga to the most expensive.
Or read how the gateway picks between them: routing and fallback, and what it costs: plans and margins.