Pareto API by Unbiased: a blended model with a 1M-token context
Pareto, the model Unbiased (Circuit & Chisel, Inc.) released on September 17, 2026, is on FastMetal as pareto.
Update, October 21, 2026: pareto now serves Pareto 26.10, released October 15. The context window grew from 262,144 tokens to about 1 million, and the price went down. The model id did not change, so callers have nothing to update.
Pareto is a blended model: Unbiased combines several underlying models behind it. To the caller it behaves like any chat model, with one request in and one answer out.
What it supports
- Context of 1,048,576 tokens, with up to 131,072 output tokens.
- Image input (text and images in, text out).
- Tool calling.
- Unbiased positions it for research, coding and agent work.
What we checked
Before listing it, and again on the switch to 26.10, we made real requests through our gateway:
- An image sent as base64 was read correctly.
- A request with tools returned a tool call.
Two things to plan for:
- Strict structured output does not hold. A strict JSON Schema in
response_formatcame back as plain prose on October 21, 2026, on both 26.9 and 26.10. If you need structured output, describe the format in the prompt and validate what comes back. - Pareto sometimes writes its reasoning straight into
contentinstead of a separate reasoning field. If you only want the answer, account for that when you parse the response.
Because the model is new, model: "auto" does not pick it for now. Ask for pareto by name to use it.
Pricing
Pareto is billed per million input tokens and per million output tokens from your prepaid balance, with no monthly fee. Current rates are on the Pareto model page and in pricing.
Calling it
The endpoint is OpenAI-compatible; pass pareto as the model:
curl https://api.fastmetal.ai/v1/chat/completions \
-H "Authorization: Bearer $FASTMETAL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "pareto",
"messages": [{"role": "user", "content": "Introduce yourself in three lines."}]
}'
Everything else we serve is in the model catalog.