Inception Artificial Intelligence Inc., a startup developing ultra-fast large language models based on diffusion, said it has raised an undisclosed amount in a funding round led by Menlo Ventures. The round also included participation from Mayfield, Innovation Endeavors, Nvidia Corp.’s venture arm NVentures, Microsoft Corp.’s M12, Snowflake Ventures and Databricks Investment. The company positions itself as a pioneer of diffusion-based LLMs and markets its Mercury model family as a commercially available dLLM.
Inception contrasts its approach with the autoregressive technique used by modern large language models, which generates tokens sequentially. The company said diffusion, the mechanism used by image and video models such as DALL-E, Midjourney and Sora, enables the model to form blocks of text concurrently rather than token by token. Inception claims this yields text generation that is up to 10 times faster and more efficient while preserving answer quality. Co-founder and chief executive Stefano Ermon described inefficient inference as the primary barrier and cost driver for deploying large-scale models and said diffusion is the path to making frontier model performance practical at scale.
The Mercury family includes a general-purpose model for ultra-low latency chat and Mercury Coder optimized for code generation. Both models offer a 128,000-token context window, roughly equivalent to a 300-page novel, and are priced at 25 cents per million input tokens with the output price unspecified in the article. Inception added a visualization feature in Mercury chat that animates text as it sharpens into detail. The company said diffusion reduces the graphics processing unit footprint required for inference, enabling larger models at the same latency and cost or supporting more users on the same infrastructure. Inception also outlined a roadmap for advanced reasoning with built-in error correction to reduce hallucinations, unified multimodal capabilities, and high-precision control over structured outputs for tool calling and data generation. Mercury is available through Amazon Web Services and unified access services like OpenRouter, with support from Microsoft Azure Foundry expected soon.
