Mistral opens its trillion-parameter model

Edward Yen

— 2 min read
Share

Saturday, 31 October 2026 · 3-minute read

Hi, and welcome to the first Lispresso. Here's what mattered in AI this week:

  • Mistral releases a trillion-parameter model anyone can download
  • OpenAI backs one robots.txt signal for AI training
  • Reddit moves toward logged-in only
  • OpenAI updates the rulebook for its models

Mistral opens its trillion-parameter model

  • Mistral has released the weights of Mistral Large 4, nicknamed "Le Chonk", so anyone can download and run it.
  • It is a 1-trillion-parameter multimodal model trained in Europe, which Mistral calls the best open-weight model for coding.
  • Teams that can't send code or security data to a closed API now have a frontier-scale model they can host themselves, but even in compressed form it is about 578 GB and vLLM runs it on four or more data-centre GPUs, so plan on serious hardware or a hosted version.

Read more: Le Chonk: Built for Coders, Trusted by Defenders. (Mistral AI); Day 0 support for Mistral Large 4 on vLLM (vLLM project)

OpenAI backs a shared signal for AI training

  • OpenAI says it will honour a shared robots.txt signal that lets website owners state their preferences on AI training.
  • The signal is shared, so one line in robots.txt speaks to every participating provider.
  • If more providers sign on, opting out of AI training gets much simpler; until then, keep your per-crawler rules, because the signal only covers providers that take part.

Read more: Making it easier for websites to express AI training preferences (OpenAI)

Reddit moves toward logged-in only

  • Reddit says it will require people to log in to use the site, as AI answers eat into the traffic it gets from search.
  • It is starting with tests outside the US.
  • When search sends fewer visitors, sites have less reason to stay open to anonymous readers, and Reddit is now testing what happens when it stops.

Read more: Reddit will force users to log in as AI is killing search traffic (Engadget)

OpenAI updates the rulebook for its models

  • OpenAI has published an updated Model Spec, its public document on how its models should behave.
  • The new version is built around five core values, including sound judgment, human agency and faithfulness to human intent.
  • If you build on OpenAI's models, the Spec is the clearest statement of the defaults you'll be working with, so read it before you write prompts that fight them.

Read more: An updated Model Spec for more capable and beneficial AI (OpenAI)

Other news you might like

That's it for today. Which of these four would you have skipped? Hit reply and tell us.

See you next time,
Lispresso

How we make Lispresso: we read public feeds from the publishers we credit, write each story in our own words and link to the original. AI tools help with drafting, and a person approves every issue before it goes out. Spot a mistake? Reply and we'll correct it in the next issue and in the archive.