Technology
October 30, 2026 1000 words

Mistral Large 4 "Le Chonk" Goes Open Weight, Plus a New Responses API and Voxtral 3

Hams AI Tech
Professional Content Team
Expert in AI Prompts & Professional Tools

Europe's leading AI lab had a busy end to October 2026. On October 29, Mistral released the weights of an improved Mistral Large 4, nicknamed Le Chonk, a one trillion parameter, natively multimodal mixture-of-experts model that had entered public preview earlier in the month. In the same window, Mistral introduced a Responses API for building agents and launched Voxtral 3, a new family of speech models. For organizations that value European hosting, open weights and competitive pricing, these releases make Mistral a serious option.

Mistral Large 4: What Improved

After three additional weeks of reinforcement learning following the preview, Mistral reports strong results, particularly in coding and agentic work:

    • Leading open-weight model on Cognition's FrontierCode 2.0 at 19.7%.
    • 41% on Terminal-Bench 4.0.
    • 48% on the Artificial Analysis Coding Agent Index, tied with GLM-5.3 while using 47% fewer tokens.
    • 51% on the Artificial Analysis Cyber Index.

The model now offers low, medium and high reasoning modes, letting teams balance speed and depth.

Pricing and Availability

Mistral Large 4 costs 0.70 dollars per million input tokens and 2.10 dollars per million output tokens through Mistral's API. That is a small fraction of what Western frontier models charge, and it places Mistral among the most cost-effective high-capability options. The model is also available on OpenRouter, Microsoft Foundry and NVIDIA NIM.

Understanding the License

The weights are on Hugging Face under the Mistral Ecosystem Open-Weight license, a modified MIT license. Most organizations can use, modify and deploy the model, but businesses offering the model as a service need a separate commercial license. If you are a law firm or consultancy deploying it internally, the standard license is likely sufficient, but confirm with counsel. Technology lawyers can analyze license terms faster with structured tools such as Claude Skills for Business Lawyers.

The Responses API for Agents

Mistral's new Responses API, in public preview, follows the OpenAI Responses format. Each response returns the model's answer, its reasoning and its tool calls as separate items. That separation makes multi-step agent loops easier to build than with the older Chat Completions style, because developers can inspect what the model decided, why and which tools it wants to use. It works with all chat models hosted by Mistral, including Mistral Large 4.

For teams building automations, this means existing agent code written for OpenAI-style APIs can often be pointed at Mistral with minimal changes. Explore our AI agent workflows and n8n workflow templates for patterns you can adapt.

Voxtral 3: Speech for Agents

Mistral also launched Voxtral 3, a set of speech models:

    • Voxtral Small Transcribe 3 and Voxtral Mini Transcribe 3 for speech-to-text, including realtime options.
    • Voxtral Mini TTS 3 for text-to-speech with emotion and pace control and voice cloning, priced at 13 dollars per million characters.
    • A voice agent SDK for building conversational applications.

All are available via Mistral's EU-hosted API or for self-deployment. For European organizations with data residency requirements, an EU-hosted voice stack is a meaningful advantage. Compare options in our voice AI roundup.

Why Mistral Matters for Professionals

Data Sovereignty

Many European law firms, healthcare providers and public bodies must keep data within the EU. Mistral offers EU hosting and open weights that can be deployed on-premises, which simplifies compliance discussions with clients and regulators.

Cost-Effective Scale

At 0.70 and 2.10 dollars per million tokens, Mistral Large 4 makes it affordable to process large document sets. Data teams can learn practical techniques from Mistral for Data Scientists: 10 Pro-Level Use Cases.

Coding and Automation

Strong coding benchmarks make Le Chonk suitable for internal developer tools. Developers can compare approaches using our software development resources and Claude Code guides.

Getting Up to Speed Quickly

If your team is new to Mistral, the Mistral Complete A to Z Guide covers setup, prompting and workflows in one place. You can also browse our Mistral section and the Mistral AI collection.

Practical Deployment Tips

    • Start with the hosted API to evaluate quality before investing in self-hosting.
    • Use low reasoning for extraction and formatting, and high reasoning for analysis and coding.
    • If you self-host, plan for substantial GPU memory: a one trillion parameter model, even as a mixture of experts, needs data center hardware.
    • Review the license terms if you plan to offer the model to customers.
    • Test Voxtral 3 transcription on your own recordings, including accents and background noise typical of your meetings.

How Mistral Compares

Mistral Large 4 competes with other large open models such as Kimi K3.1 Pro and Meta's Muse Spark Open. Mistral's distinguishing features are its European base, EU hosting, aggressive pricing and a coherent stack of text, agent and voice products from one provider. For organizations that want to reduce dependence on US and Chinese providers, that combination is attractive. It also simplifies vendor management: one contract, one data processing agreement and one support relationship can cover text, agents and voice.

Frequently Asked Questions

Is Mistral Large 4 truly open source?

The weights are openly available under a modified MIT license. Model-as-a-service businesses need a separate commercial license, so it is best described as open weight with conditions.

Where can I use Mistral Large 4 besides Mistral's API?

It is available on OpenRouter, Microsoft Foundry and NVIDIA NIM, as well as through self-deployment from the Hugging Face weights.

What does the Responses API add?

It separates answers, reasoning and tool calls into distinct items, which makes agent loops easier to build and debug.

Does Voxtral 3 support voice cloning?

Yes. Voxtral Mini TTS 3 includes voice cloning along with emotion and pace control. Always obtain consent before cloning anyone's voice.

The Bottom Line

Mistral now offers a credible full stack: a frontier-scale open model, an agent-friendly API and modern speech models, all available from Europe at low cost. For privacy-conscious professionals and budget-aware teams, it deserves a place on the shortlist. Explore our technology AI resources to start building.

Related AI News

Topics & Keywords
Mistral Large 4 Le Chonk Mistral AI open weights Voxtral 3 Responses API European AI