Mistral Large 4 Goes Open-Weight, Plus a New Radar Skill

The conversation

AI-generated episode: the script is written with AI from the cited sources and read by synthetic voices; the hosts are fictional.

Sveva

Welcome to Hot Podcast! It's the thirtieth of October, and a trillion-parameter model just landed in the open. Stefano, Matteo, ready to dig in?

Stefano

Ready! A trillion parameters. My laptop just quietly asked to be left out of this conversation.

Sveva

Here's the news. TestingCatalog reported on 29 October 2026 that Mistral released the weights of Mistral Large 4, a natively multimodal flagship, for coding, cybersecurity and enterprise agents. It has Low, Medium and High reasoning modes, and a modified MIT license permitting free use, modification, fine-tuning and commercialization.

Matteo

Where can people actually get it? That's usually my first question.

Stefano

The article lists Mistral Studio, the Mistral API, OpenRouter, Hugging Face, Microsoft Foundry and NVIDIA-hosted endpoints. You can also self-host with vLLM or SGLang. It also cites Surge Code blind comparisons placing it behind only Claude Opus 5.5 among tested models. That's the report, not my own testing.

Stefano

Now, my analysis. The reasoning modes could let teams tune cost against performance. And open weights might give security-minded companies more control over where the model runs. But a trillion parameters likely means serious hardware, so self-hosting won't suit everyone.

Matteo

Agreed. Hosted endpoints may be the easy first step. Also, this is secondary coverage, not Mistral's own announcement, so I'd check the original before building anything important.

Sveva

Speaking of keeping up, there's a new skill to explore on HotSkills: ai-listening, from the new-term-radar repository. It's a separate discovery, not tied to this story.

Matteo

Its description says it collects what's new in AI from the last 24 hours across X, Hugging Face, GitHub and sitemaps, deduplicated, into one report. It runs a command line tool and needs an AISA_API_KEY. The repository's heat metric was 9.5 on 30 October 2026, which is repository heat, not a vote.

Stefano

Purely hypothetical: on a launch day, a report like that could surface fresh model and repository names for a team's research. Though the skill itself has a section on what it cannot tell you, so it's no crystal ball.

Matteo

And it flags cost and time upfront. I like tools that say the awkward part first. Unlike my gym membership.

Sveva

So the takeaway: open weights widen the options, but check the original source, your hardware and the license details. And treat any new tracking tool as a starting point, not the final word.

Sveva

You'll find the episode sources on HotSkills. Thanks to Stefano and Matteo, and thank you for listening. Until next time!