01What Is Mistral Large 4?

Mistral Large 4 is a new AI model from the French company Mistral. It is also called Le Chonk. This is a 1 trillion parameter model that works with text and images.

It uses a mixture of experts design, which means only part of the model activates at a time. Mistral claims it is the best open weights model from the US or Europe on benchmarks.

The model was trained from scratch on 3800 Nvidia Grace Blackwell GPUs in Europe. It has a 500,000 token context window. That means it can handle very long documents or conversations.

Mistral Large 4 is designed to be state-of-the-art in areas like cybersecurity and manufacturing. You can access it right now through Mistral's API or through OpenRouter.

02Main Features of Mistral Large 4

  • 1 trillion parameters with 52 billion active parameters per use.
  • Multimodal: works with both text and images.
  • Mixture of experts architecture for efficiency.
  • 500K token context window for long inputs.
  • Focus on cybersecurity and manufacturing tasks.
  • Open weights (coming soon) so developers can run it themselves.
  • Available as a public preview on Mistral API and OpenRouter.
  • Pricing: $0.68 per million tokens input, $2.09 per million tokens output.

03How to Access Mistral Large 4

You can try Mistral Large 4 right now. There are two main ways to get started.

  1. Go to Mistral's studio and sign in. Apply for API access. This is a paid plan.
  2. Use OpenRouter. Search for Mistral Large 4. You will see the price and can start using it immediately.

OpenRouter lets you compare models side by side. This is what Julian used in the episode to test it against Kimi K3. The open weights are not released yet, but Mistral says they will come soon. Public preview is available now.

04How Mistral Large 4 Performed in Tests

Julian tested Mistral Large 4 against Kimi K3. He gave both models the same task: create a beautiful website for the AI profit board and community about AI automation. Kimi K3 finished quickly.

It produced a usable HTML website that looked decent for a first attempt. Mistral Large 4 was still thinking after Kimi K3 finished. It did not return any output during the test.

Julian also tried Mistral Large 4 on another task earlier that day. That task was also still running. The model seemed very slow on OpenRouter.

He noted that this might be because of the OpenRouter API, not the model itself. But even so, he found it less reliable than other open source models. Kimi K3 and GLM 5.3 both performed better in speed and output quality.

05Benchmarks and Comparison Results

According to the Artificial Analysis index, Mistral Large 4 scored 38 overall. That puts it below DeepSeek, GLM 5.3 Flash, Kimi K3 Max, GLM 5.3 Max, and MIMO V2.6 Pro. Chinese models still lead the open weights leaderboard.

Mistral Large 4 performed best in cybersecurity. It was on par with GLM 5.3 on that task. On agentic behavior, it was slightly below GLM 5.3.

Its grounding score was 42, just above GP6 Astra at 41.5.

On coding tasks, GLM 5.3 still beat Mistral Large 4. Overall, it did not top any of the major categories except cybersecurity, where it tied. Julian concluded that Mistral Large 4 is not yet competitive with the best open models.

He would not switch to it.

06Tips, Limits, and Mistakes to Avoid

  • Do not expect Mistral Large 4 to be faster than Kimi K3 or GLM 5.3. In tests, it was much slower.
  • Do not rely on it for tasks that need quick output. It may leave you waiting.
  • If you use OpenRouter, the model might be slower than on Mistral's own API. Test both before deciding.
  • Mistral Large 4 is not yet open weights. You cannot run it on your own hardware yet. Wait for the open release if that matters to you.
  • For most practical use cases, other models like Kimi K3, GLM 5.3, or Claude are more reliable.

07Who Should Use Mistral Large 4

Mistral Large 4 is for developers and researchers who want to test a new European open source model. It is especially interesting if you work in cybersecurity. That is the one area where it matches top models.

If you need a fast, reliable model for general tasks, Mistral Large 4 is not the best choice right now.

Julian says it is a good idea that Europe is building open source models. But he has covered Mistral since launch. He was not impressed.

For daily work, he recommends Kimi K3 or other models. Mistral Large 4 matters because it shows Europe is investing in AI. But the results so far are not strong enough to replace what you already use.