Artificial Atlas

Who is betting on what in AI, and against whom.

Inception

Diffusion language models: text generated by denoising blocks in parallel, not token by token.

HQ
Palo Alto
Founded
2024
Web
inceptionlabs.ai
Camps
Other substrates LLM scaling

Edit this entry

What it does

Stefano Ermon, a co-inventor of the diffusion methods behind image generation, with former students Aditya Grover and Volodymyr Kuleshov. Mercury, launched February 2025, generates text by iteratively denoising blocks of tokens, which the company reports at more than 1,000 tokens per second on one H100. Same training data as camp 1, a different generative process, which is why it sits with the substrates.

Camp membership over time

  1. dates not recorded
    Program search and other substrates
    Diffusion instead of autoregression for language
    Sources Inception raises $50 million to build diffusion models for code and text TechCrunch 2025-11-06 · Mercury: Ultra-Fast Language Models Based on Diffusion arXiv 2025-06-17
  2. dates not recorded
    Scale next-token prediction (secondary)
    Trained on the same text corpora and sold against camp 1 models
    Sources Mercury: Ultra-Fast Language Models Based on Diffusion arXiv 2025-06-17

Capital

AmountKindValuationDateInvestorsSource
$50Mseed Nov 2025 Menlo Ventures, Mayfield, Innovation Endeavors, M12, Snowflake Ventures, Databricks, NVentures TechCrunch

People

No people recorded at this lab yet.

General sources

Inception raises $50 million to build diffusion models for code and text TechCrunch 2025-11-06