Inception Raises $50M to Advance Diffusion AI Models and Launch Mercury

As stated by Techcrunch

Indeed, Inception raised $50 million in a seed round led by Menlo Ventures. Additional angel funding was provided by Andrew Ng and Andriy Karpaty.

The project’s lead is Stanford University professor Stefano Ermon, whose work focuses on diffusion models that generate results through iterative refinement rather than step-by-step word prediction. Such models underpin AI image-processing systems like Stable Diffusion, Midjourney, and Sora. Drawing on the experience building these systems, Ermon aims to apply the same approach to a broader range of tasks.

Along with the funding, the company released an updated version of Mercury – a model designed for software development. Mercury has already been integrated into a number of development tools, including ProxyAI, Buildglare, and Kilo Code. Most importantly, according to Ermon, the diffusion approach helps reduce latency and computational costs.

“These diffusion-based LLMs are significantly faster and more efficient than anything else being built today,”

– Stefano Ermon

“It’s simply a completely different approach, with a lot of innovations that can still be brought to the table.”

– Stefano Ermon

Ermon also notes that their system already demonstrates speeds of over 1,000 tokens per second, far exceeding the capabilities of existing autoregressive technologies, since our solution is designed for parallel processing and is exceptionally fast.

Moreover, diffusion models provide greater flexibility in hardware usage: unlike traditional autoregressive models, they allow handling many operations concurrently, reducing overall latency in complex tasks.

In the future, Inception plans to broaden the application of Mercury and continue developing diffusion models for various tasks – from development tools to large language systems – with a focus on reducing computing costs and response times.

Prospects and Future Steps

The company is exploring using Mercury for a wider range of tasks and will continue refining diffusion-model algorithms for text and code to make them more accessible to industry and developers.

Don’t miss other news: