Inception Labs

Inception Labs is a small American company that bet on an unusual architecture: instead of models that write one word after another, it builds diffusion language models, which refine a whole draft in parallel. Its models are sold under the Mercury name and are priced aggressively low per token.

The company’s pitch is not that it beats the frontier labs on intelligence, but that it wins on throughput and cost for work where speed is the bottleneck: code completion, voice assistants, and processing large batches of documents.