What will it cost?
Estimate a request, compare models and plan a month of use. No sign-in needed.
Set your request size
Your prompt and system prompt never leave this browser on this page. The calculator fetches the public model catalogue and computes costs here. Shared links contain choices and token counts, never prompt text.
Add a system prompt (optional)
100 estimated input tokens · 512 output tokens · 3,000 requests per 30-day month
Token counts are estimates: about one token per four characters, using the same helper as the Harness. Presets use 100 input tokens for a question, 1,500 for a coding task, or 500 per document page (about 2,000 characters). Actual token use depends on the model, language and request format.
Compare live models
What the tags mean
- Reads images. Accepts images alongside text; image input is declared by the provider.
- Makes images. Returns images; catalogue rates may include token and image charges.
- Audio. Accepts or returns audio; browser speech works separately from model audio.
- Tools. Can request tool calls; the calling app must run the tools and return their results.
- Long context. Lists a context window of at least 128,000 tokens; the chosen provider may have a lower limit.
- Proven hardware. Has an endpoint with a fresh hardware attestation checked by the router; this does not prove answer quality or encrypt ordinary chat.
- AnyRoute network. Has a live offer on an AnyRoute network host for a model admitted on that host; routing may choose another provider. Read more →
- Encrypted chat. Available through the separate device-encryption gateway setup; ordinary Harness chat still sends readable requests to the router. Read more →
Tags describe available endpoints. A request’s lane, provider and signed receipt show which path was used.
Prices from the live catalogue; routing may pick another provider
USD estimates use input and output token rates plus any listed per-request fee. They exclude image, audio, search, cache and separate reasoning charges, conversation history and request-format overhead. They do not reserve a price or check whether your request fits a model’s limits. Amounts round half up to six decimals per request and two per month; monthly totals use the unrounded request cost. Small positive amounts are shown with <.
Loading the live catalogue…