Mistral Large 4: the 1.05T-parameter open model they call Le Chonk
Announced October 6, 2026, Mistral Large 4 (ML4) is a Mixture-of-Experts model with 1.05 trillion total parameters and 49 billion active per token, a 1-million-token context window, and native multimodal support. "Le Chonk" is the official nickname Mistral used in its own launch announcement. This guide collects the announced specs, benchmarks, preview API pricing, and the expected October 27, 2026 open-weights release in one place.
API preview
Live
Launched October 6, 2026 via Mistral's API.
Open weights
Expected Oct 27, 2026
Reuters reported October 27, 2026. Expected on Hugging Face under a custom Mistral license.
Preview API pricing
$1.36 / $4.18
$1.36 per 1M input tokens, $4.18 per 1M output tokens.
Weights release countdown
Targeting the reported October 27, 2026 open-weights release.
Explore the guide
What is Le Chonk
Mixture-of-Experts architecture, the 1.6B-param vision encoder, training on 3,800 Nvidia Grace Blackwell GPUs, and the people behind ML4.
Benchmarks
93% Cybench and 82% CyberGym-E2E as reported by Mistral, plus the Artificial Analysis leaderboard placement.
vs DeepSeek V4 Pro
How ML4 compares to DeepSeek's flagship, with the ML4 preview sitting above DeepSeek V4.1 Flash on Artificial Analysis.
vs Qwen3.8 Max
Head-to-head on parameters, context window, multimodality, and open-weights availability.
vs closed frontier models
ML4 against entry-level closed models like GPT-6 Luna on capabilities and open-weights status.
API pricing & calculator
The $1.36/$4.18 per 1M tokens preview pricing, and a calculator for estimating your API spend.
Download & weights
Where the weights are expected to land on October 27, 2026, under a custom Mistral license.
Quickstart
Get up and running with the ML4 API preview, from API access to a first request.
Changelog
A running log of announcements and changes from the October 6, 2026 launch onward.
FAQ
Answers to common questions about Le Chonk's specs, pricing, license, and weights release.