Unbiased model releases

Pareto is the first public model from Unbiased (built by the Circuit & Chisel team), released 2026-09-17 after debuting anonymously on OpenRouter as the "Union Alpha" stealth model. It is a multimodal "composite" model built for research, coding and agentic workflows: rather than a single set of weights, Pareto blends several frontier and open-source models — routing each request across them and returning the best result at an efficient price. It accepts text and image input and returns text output, with a 262,144-token context window (max output 131,072 tokens), and supports function calling, tool choice, structured outputs and prompt caching. It is served as a proprietary, hosted-only model (no open weights) through Unbiased's own endpoint (and via OpenRouter as unbiased/pareto) at $2.50 / 1M input and $7.50 / 1M output (cache read $0.25). Because it is an ensemble/router over other models, its underlying architecture and size are undisclosed. Unbiased positions it at "frontier-level performance across a broad range of general-purpose tasks"; performance claims are vendor / self-reported.