Inception Labs AI, a Palo Alto–based company pioneering diffusion-based large language models (dLLMs), has raised $50 million in fresh funding. The round was led by Menlo Ventures, with participation from Mayfield, Innovation Endeavors, NVentures – NVIDIA’s venture arm, Microsoft’s M12, Snowflake Ventures, and Databricks Investment.
The funding will accelerate product development, expand research teams, and enhance real-time AI capabilities across text, voice, and code.
The challenges it tackles
Most large language models today rely on autoregression, a method that generates words one at a time. While accurate, this sequential process is painfully slow and costly, making it difficult for enterprises to scale or deliver seamless real-time experiences.
Inception takes a radically different approach. Its diffusion-based LLMs (dLLMs) borrow principles from image and video generation technologies such as DALL·E, Midjourney, and Sora. Instead of crafting text word by word, dLLMs generate entire responses in parallel. This innovation delivers outputs that are 10x faster and far more efficient, without sacrificing coherence or accuracy.
Mercury: Speed meets intelligence
Inception’s flagship model, Mercury, is the world’s first commercially available diffusion LLM. It outpaces even speed-optimised models from OpenAI, Anthropic, and Google by 5–10x, while matching their precision. Mercury comes in two versions: a general-purpose model for conversational tasks and Mercury Coder, tuned for code generation. Both feature a 128,000-token context window, equivalent to about 300 pages of text, allowing for deep and complex interactions.
By drastically cutting GPU requirements, Inception enables organisations to run larger models at the same cost or serve more users using existing infrastructure. This makes Mercury ideal for latency-sensitive applications such as live voice assistants, dynamic UIs, and real-time programming tools, areas where speed defines usability.
Inventors of a new language era
Inception Labs AI was co-founded by Stefano Ermon, Aditya Grover, and Volodymyr Kuleshov in 2024 in Palo Alto, California. The founding trio, who have roots at Stanford, UCLA, and Cornell, were among the early researchers behind core AI advances such as diffusion, flash attention, and direct preference optimisation. CEO Stefano Ermon is also a co-inventor of the diffusion techniques that power systems like Midjourney and OpenAI’s Sora.
Beyond speed, Inception is developing models with built-in error correction to reduce hallucinations, unified multimodal capabilities to handle language, image, and code seamlessly, and structured output control for precise tasks like data generation. With backing from top investors and availability via Amazon Bedrock, OpenRouter, and Poe, Inception is redefining how intelligent systems think, speak, and respond in real time.
“The team at Inception has demonstrated that dLLMs aren’t just a research breakthrough; it’s a foundation for building scalable, high-performance language models that enterprises can deploy today,” said Tim Tully, Partner at Menlo Ventures. “With a track record of pioneering breakthroughs in diffusion models, Inception’s best-in-class founding team is turning deep technical insight into real-world speed, efficiency, and enterprise-ready AI.”
“Training and deploying large-scale AI models is becoming faster than ever, but as adoption scales, inefficient inference is becoming the primary barrier and cost driver to deployment,” said Inception CEO and co-founder Stefano Ermon. ”We believe diffusion is the path forward for making frontier model performance practical at scale.”
