Model prices

How to read this. We charge a flat 1.4× markup on the provider's published rate, on every model — the same multiplier everywhere, so a cheaper model is genuinely cheaper per message. Cached input is billed at about 1/50 of the input rate. Off-peak (weekends, and weekdays outside 12:00–17:59 UTC) is roughly half price on models that publish a night rate. Your plan's monthly allowance and your balance are both spent at these prices, per token actually used.

Typical message assumes 25 000 input tokens and 1 000 output tokens — roughly a long document question or a page of code. A short chat turn costs far less. Prices come from the provider's own published page, refreshed daily at 05:00 UTC; this table is generated from that catalogue at load time, never typed by hand.

1 · WebX11 cloud models — what the table prices. Metered per token from your plan's monthly allowance, then from your balance, at the 1.4× retail rate. No key to manage, works in ChatStudio, newOS4 and the API.

2 · Local models on this server — llama3.2, hermes3, codecraft, webx11-devcourse and the rest of the local set run on this host and cost $0. Free within fair use: they are metered and logged like everything else (so we can see load), but priced at zero, and they never touch your allowance or balance. Slower than the cloud models and weaker on very long context — that is the trade.

3 · Your own provider — in newOS4 you can add any Ollama- or OpenAI-compatible endpoint with your own API key (providers: {name, type, endpoint, apiKey, model}). Those calls go straight to that provider: you pay them directly at their price, and they are not metered, priced or billed by us. Use this if you already have credits elsewhere or need a model we do not carry.