Mistral released the weights for Mistral Large 4, a trillion-parameter open-weight model it is promoting for cybersecurity and coding. Organizations can evaluate it in their own environments, but should review the license, security controls, deployment requirements, and independent performance results before using it with sensitive workflows.
Key takeaways
- Mistral Large 4, also called Le Chonk, is reported to have 1.05 trillion total parameters and 52 billion active parameters.
- The model supports a context window of up to 1 million tokens, according to reported specifications.
- Mistral promotes ML4 for coding and cybersecurity, but its benchmark and efficiency statements are company claims that require independent validation.
- Open weights can support self-hosting, but they do not automatically make a deployment secure or mean the model is open source.
- Before testing ML4 with sensitive work, review its custom license, infrastructure needs, data handling, access controls, and performance on representative tasks.
Mistral has released the weights for Mistral Large 4 (ML4), a trillion-parameter model it is positioning for cybersecurity and automated coding. The October 29 release gives organizations the option to run the model in their own environments rather than rely only on an external AI service - a potential advantage for teams handling sensitive code, security data, or government workloads.
The release makes the Mistral cybersecurity model available for evaluation, but Mistral’s performance claims still need to be tested against independent results and an organization’s own security requirements.
What Mistral Released: A Trillion-Parameter Open-Weight Model
Mistral Large 4, also called “Le Chonk,” has 1.05 trillion total parameters, with 52 billion active parameters, according to reported specifications. It supports a context window of up to 1 million tokens, allowing it to process large amounts of text in a single interaction.
The release is described as open-weight, which means organizations can access model weights for deployment and adaptation. It is not necessarily open source: reporting describes the model as carrying a custom license, so teams should review the terms before using it commercially or modifying it.
What Mistral Claims About Cybersecurity and Coding Performance
Mistral presents ML4 as a leading open-weight model for automated coding and cybersecurity. The company says it ranks above other open models on Artificial Analysis’ Coding Agent Index. It has also claimed that Le Chonk is the world’s best open-weight model for these tasks.
Those are company claims, not proof that the model will outperform alternatives in every security workflow. Benchmark scores can provide a comparison point, but results depend on the tasks, prompts, tools, and evaluation methods used. Security teams should test performance on representative work, such as code review or vulnerability triage, and check outputs for accuracy and unsafe recommendations.
Mistral has also compared ML4 with GLM-5.3, describing performance as comparable while using 47 percent fewer tokens per task. VentureBeat separately reported a claim of 63 percent lower cost per task. These figures may offer a reason to investigate efficiency, but actual costs and results will vary with deployment choices and workload.
What Self-Hosting Could Mean for Security Teams
Running an open-weight model in an organization’s own environment can offer more control over where data is processed and how the model is integrated. For companies and government agencies, that may reduce dependence on external AI providers and make it easier to align deployment with internal data-handling rules.
Self-hosting does not automatically make a system secure. Teams still need to assess infrastructure, access controls, logging, model updates, and the risks of connecting an AI agent to code repositories or security tools. A model that can assist with cybersecurity tasks may also produce incorrect advice or require safeguards against misuse.
Where to Get the Weights and What to Check First
The model is available through Hugging Face and Mistral’s API, according to reporting on the release. Before a trial, organizations should check the custom license, hardware and operating requirements, and how the chosen deployment handles sensitive data. They should also compare independent benchmark results with tests on their own tasks.
For security teams, ML4’s weight release is a new option for controlled deployment, not a guarantee of better security outcomes. Review the license, deployment requirements, and independent results before testing it with sensitive workflows.
By the numbers
1.05 trillion total parameters and 52 billion active parameters
These are reported specifications for Mistral Large 4, also called Le Chonk; the figures are not presented here as independently verified benchmark results.
A context window of up to 1 million tokens
Reported model specifications describe this as the maximum context window, allowing the model to process a large amount of text in one interaction.
47% fewer tokens per task than GLM-5.3
Mistral made this comparison for ML4; actual token use may vary by workload, prompts, and deployment, so the claim should be independently tested.
63% lower cost per task
VentureBeat separately reported this as a claim comparing ML4 with GLM-5.3; actual costs will depend on workload and deployment choices.
Step by step
- 01
Review the model license
Read Mistral Large 4's custom license and confirm whether your intended commercial use, modification, and deployment are permitted.
- 02
Check deployment requirements
Assess the hardware and operating requirements, and determine where data will be processed and stored in your chosen deployment.
- 03
Set security controls
Define access permissions, logging, update procedures, and safeguards for any connections to code repositories or security tools.
- 04
Test representative workflows
Evaluate the model on tasks such as code review and vulnerability triage, checking accuracy and unsafe recommendations against your own criteria.
- 05
Compare independent results
Review independent benchmark evidence alongside your test results, and treat vendor performance and cost comparisons as claims to verify.
Frequently asked questions
What is Mistral Large 4?
Mistral Large 4 is an open-weight AI model that Mistral is promoting for coding and cybersecurity. Reported specifications describe it as having 1.05 trillion total parameters, 52 billion active parameters, and a context window of up to 1 million tokens. It is also known as Le Chonk.
Can organizations self-host the Mistral cybersecurity model?
Organizations can use the released weights to evaluate self-hosted deployment, subject to the model's license and technical requirements. Self-hosting can provide more control over data processing and integration. It does not by itself secure the infrastructure or prevent inaccurate model outputs.
Is Mistral Large 4 open source?
The article describes Mistral Large 4 as open-weight, not necessarily open source. Reporting says it has a custom license, so organizations should check its terms before commercial use, modification, or deployment. Access to model weights alone does not establish that all source materials or usage rights are unrestricted.
How should security teams evaluate Mistral Large 4?
Security teams should test Mistral Large 4 on representative tasks and review its license, deployment requirements, data handling, and safeguards. Use cases such as code review and vulnerability triage can reveal whether its outputs meet the team's accuracy and safety requirements. Compare internal results with independent benchmarks rather than relying only on vendor claims.
