Hugging Face releases Carbon-X, a 14.8B-parameter model for reading DNA

⚡ The gist in 5 seconds - Hugging Face has released Carbon-X — an open language model with 14.8 billion parameters for reading and analyzing DNA. - The announcement was made by the company's co-founder and CEO Clément Delangue on October 22, 2026, via his @ClemDelangue account. - The main talking point around the release is the debate over how censorship filters in closed models hinder science and development. ### 🔍 What was found On October 22, 2026, Clément Delangue, co-founder and CEO of Hugging Face, announced the release of Carbon-X — an open language model with 14.8 billion parameters. The model works with DNA: it reads and analyzes sequences. The statement was made via his @ClemDelangue account. The analysis on Habr places the release at the center of a broader discussion — about censorship in closed AI models and how it affects scientific work and development. A separate thread of the debate concerns "shallow filters": restrictions that cut off a request without understanding its substance. The mechanics of the filters are not disclosed in the announcement, and the analysis is built around the Carbon-X case. ### 💡 Why it matters Open weights give researchers something closed-model APIs don't: the model can be downloaded, run on your own data, and studied from the inside. For bioinformatics, this is a sensitive topic — queries about DNA and genomes often run into safety filters written for general cases. The Carbon-X case highlights a fork in the road: either a scientific problem is solved on an open model, or it gets blocked by someone else's moderation policy.