Key Notes
- Mistral has released downloadable Le Chonk weights following its October preview, allowing organizations to customize and deploy the model on their own infrastructure.
- The company reports coding performance roughly tied with GLM-5.3 on a specific index, while leading Chinese models remain ahead on broader aggregate intelligence tests.
- API rates are $0.70 per million input tokens and $2.10 for output, and model-as-a-service businesses face a separate commercial-license requirement.
Mistral AI has released the weights of Large 4, its trillion-parameter flagship nicknamed Le Chonk, giving developers the option to run and customize the model on their own infrastructure. The French company is taking aim at coding agents and cybersecurity, with API pricing now listed at $0.70 per million input tokens and $2.10 for output.
The October 29 release fulfills the promise made during the model’s October preview. Hugging Face detailed the downloadable checkpoints and software support in its release guide, while Mistral says additional training has improved coding performance and reduced the number of tokens needed to complete tasks.
Coding Performance Takes Aim at GLM-5.3
In its latest announcement, Mistral reports a 48% score on the Artificial Analysis Coding Agent Index, describing the result as effectively tied with GLM-5.3 from Z.ai. It says Large 4 used 47% fewer tokens per task on average in that comparison, a potentially important difference for developers paying for repeated agent runs.
The company also reports 41% on Terminal-Bench 4.0, an improvement of 13 percentage points over the preview, alongside a 30% reduction in output tokens. Those figures concern particular coding evaluations and configurations; they do not establish that Le Chonk matches every Chinese model across all tasks.
The broader picture is more measured. Independent evaluator Artificial Analysis lists the high-reasoning version at 26 on Intelligence Index v5. Mistral’s model card describes that as the highest score among Western open-weight models, while acknowledging that leading Chinese open models score higher on the aggregate index. The release strengthens Europe’s offering without settling the wider AI race.
Open Weights Give Businesses More Control
The model card describes a mixture-of-experts system with roughly one trillion parameters, 49 billion active per token, or 52 billion when embeddings and output layers are included. It accepts text and images, produces text and supports a one-million-token context window. That combination is intended for work spanning code, documents, diagrams and multi-step tool use.
Official FP8 and more compact NVFP4 checkpoints are available, alongside community GGUF versions. Hugging Face lists support across Transformers, vLLM and llama.cpp. Availability of the weights lets a team choose its deployment and adapt the system, although running a model of this scale still requires substantial hardware; downloading it does not remove operating costs.
There is also a commercial condition. The license permits use and modification but requires businesses operating model-as-a-service offerings to obtain a separate commercial license before commercial use. It explicitly exempts internal use and use through Mistral’s official products from that requirement. Calling the release open-weight is more precise than implying that every deployment is unrestricted.
API Prices and Defensive Security Are Central to the Pitch
Mistral’s current pricing page lists $0.70 for a million input tokens, $0.07 for cached input and $2.10 for output. As an illustration, one million uncached input tokens plus one million output tokens would cost $2.80 at those rates. A complete agent task can make several requests, so its final bill depends on usage across the whole workflow.
Security is another major selling point. Mistral reports 51% on the Artificial Analysis Cyber Index and says the model can reproduce and patch vulnerabilities while distinguishing legitimate defensive requests from malicious ones. These are evaluation results, not a guarantee of safe behavior in every deployment. Teams using the model against real systems still need authorization, access controls and review before changes go live.
That strategy builds on Mistral’s earlier security push for organizations seeking alternatives to restricted models. With the weights now available, customers can move beyond testing a hosted preview to evaluating a deployment they operate themselves. The deciding factors will include patch quality, coding reliability and total cost on their own workloads.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
