Mistral AI's "Le Chonk" — a natively multimodal, ~1.05-trillion-parameter mixture-of-experts model that reads text and images, writes text, handles 160+ languages, and is headed for an open-weight release.
Last verified · 2026-10-06 · by Moe Ameen
Mistral Large 4, nicknamed "Le Chonk," is the European lab Mistral AI's frontier language model, announced on October 6, 2026. It is a natively multimodal mixture-of-experts (MoE) model: roughly 1.05 trillion total parameters, of which only about 49 billion activate for any given token, paired with a 1.6-billion-parameter vision encoder so it can read images as well as text. That MoE design is how a model this large stays affordable to serve. Mistral says it was trained on 3,800 NVIDIA Grace Blackwell GPUs in its own datacenters in Europe.
It launched in public preview through Mistral's API, and Mistral has said it will release the open weights later in October 2026 — pitching it as the strongest open-weight model developed in the US or Europe — though it has not confirmed the license. It supports more than 160 languages, including every official language of the European Union, and lists a context window of up to one million tokens. Some third-party hosts serve a smaller window, so treat 1M as the vendor ceiling rather than a guaranteed served length.
On Mistral's own numbers, Large 4 posts strong coding and agentic results (for example 61.7% on DeepSWE v1.1, 93% on Cybench, and a 49.8% Coding Agent Index it says edges DeepSeek V4 Pro and Qwen3.8 Max) along with knowledge-work and visual-grounding scores it claims rival frontier closed models. As with any vendor-reported launch benchmarks, independent evaluation will take time to settle where it truly lands.
The honest framing for a creator: Large 4 is a reasoning-and-language brain. It is multimodal on input and text-only on output — it reads images and text and writes text. It generates no images, video, or audio, and it publishes nothing.
Use Mistral Large 4 as your writing room. Its real edge for a creator is the front of the pipeline: dump an hour-long transcript, a stack of research, or a quarter of notes into its long context window and have it return a tight script, the three sharpest angles, and — because it speaks 160+ languages — the same script adapted for every market you publish in. That solves *what to say*, in *which languages*. It does not make a single piece of media, and it cannot put anything in front of an audience.
That is the hand-off to [Kompozy](/). Take the script Large 4 wrote — or the localized set of scripts — and Kompozy generates the finished formats a text model never could: a [Persona Short](/glossary/persona-shorts) narrated by a face-locked HeyGen avatar, [Clipped Shorts](/) from a longer talk, a brand-exact [Carousel via HyperFrames](/glossary/hyperframes), Photo Posts, Quote Graphics, a Blog Article, and an Email Newsletter — eighteen formats in all, each rewritten to your voice through the [Persona Brief](/glossary/persona-brief). Then [Autopilot](/glossary/autopilot) reframes every clip to 9:16, 1:1, and 16:9 and schedules and publishes across eight social platforms plus blog and email behind a per-post review. For a multilingual creator the pairing is especially sharp: Large 4 produces one idea in ten languages, and Kompozy turns each into a localized video-and-carousel set and ships it to the right market. Large 4 is the brain that drafts and localizes; Kompozy is the engine that renders and distributes.
Mistral Large 4, nicknamed "Le Chonk," is a natively multimodal mixture-of-experts language model from the European lab Mistral AI, announced October 6, 2026. It has roughly 1.05 trillion total parameters (about 49 billion active per token), reads text and images, writes text, supports 160+ languages, and launched in public preview via Mistral's API with open weights promised for later in the month.
No. It is multimodal on input — it can take images and text — but its output is text only. It does not generate images, video, or audio. To turn its scripts into finished video, carousels, or graphics you pair it with a content engine such as Kompozy, which renders the media and publishes it.
It launched October 6, 2026 as a paid public preview on Mistral's API; Mistral says it will release open weights later in October but had not confirmed the license, so "open-weight" is the safer term. API pricing started around $1.36/$4.18 per million input/output tokens and was reduced on the model card to roughly $0.68/$2.09. Check Mistral's current pricing before budgeting.
The front of the workflow: scripting and summarizing from long inputs thanks to its large context window, and multilingual drafting and localization across 160+ languages. It is a drafting and localization brain, not a media maker — the production and publishing happen in a tool like Kompozy.