Particle.news
Download on the App Store

Inception Raises $50 Million to Scale Diffusion LLMs for Code and Text

The startup says its Mercury models generate in parallel for faster, lower‑cost responses than autoregressive systems.

Overview

  • Menlo Ventures led the seed round, with participation reported from Mayfield, Innovation Endeavors, NVentures (NVIDIA), Microsoft’s M12, Snowflake Ventures, and Databricks Investment, and angel backing from Andrew Ng and Andrej Karpathy.
  • Inception’s Mercury lineup includes a general model and Mercury Coder for software work, both described as supporting a 128,000‑token context window.
  • CEO Stefano Ermon says internal and partner tests show throughput over 1,000 tokens per second due to the models’ parallel generation architecture.
  • A new Mercury release focused on software development was announced, with integrations into tools such as ProxyAI, Buildglare, and Kilo Code.
  • The company says the funding will expand research, product development, and real‑time capabilities across text, voice, and code, with availability listed via Amazon Bedrock, OpenRouter, and Poe.