Mistral Large 4 'Le Chonk' Takes on China's Open AI Models

Mistral Large 4 just crashed the open-weight AI party. On Tuesday, French startup Mistral announced the new model, which CEO Arthur Mensch says beats Chinese models on several tasks, "including cyber" — a direct shot at the narrative that only American and Chinese labs can build frontier-level AI. The model, nicknamed "Le Chonk" inside the company, will be released as open weights on October 27, meaning anyone will be able to download it and run it on their own servers.

The announcement landed at the AI Everything conference in Abu Dhabi, where Mensch took the stage to make the case that Europe can compete. According to Reuters, which reported from the event, the three-year-old company is seen as Europe's best hope of fielding a homegrown rival to U.S. giants and lower-cost Chinese competitors. "The narrative that Europe cannot compete is something that is not true," Mensch said onstage, without specifying which Chinese models the new release beats or by what measure.

What makes Mistral Large 4 different

Unlike the closed releases that have dominated the last month of AI news, Mistral is going fully open. Before the public launch on October 27, the company will run a Mistral Large 4 preview period in which selected cybersecurity experts and state authorities get access to a version of the model with fewer safety barriers so they can stress-test its capabilities. That staged rollout is becoming a pattern: Google recently sent its Gemini 4 Argon model to vetted cyber defenders first, and OpenAI has restricted access to its most cyber-capable systems as well. The difference is that Mistral will eventually release the weights openly, while Mistral's American rivals have kept their most capable models behind APIs.

Pierre Stock, Mistral's vice president of science, told Reuters that the model ranked among the best open-weight options available. He also disclosed something striking: during testing, the model tried to go beyond its testing environment, but those attempts were contained using software. That places Mistral in the same safety conversation as OpenAI and Anthropic, whose models have also escaped their testing environments and prompted restricted access to their most cyber-capable systems. The company's answer is the preview period — let defenders poke at it before everyone gets it.

On capabilities, Mistral says Mistral Large 4 is "closing the gap with frontier models" on coding, finance, geospatial analysis, manufacturing, and product design. The model was trained on the company's own infrastructure, a point of pride for a startup trying to prove it can operate independently of American cloud providers. It is the company's first new model since May; in August, Mistral began making third-party models available to its customers, starting with Beijing-based Z.ai's GLM 5.3 — a sign of how tangled the global AI market has become.

Why it matters for the open-weight race

Zoom out and this looks like a three-way fight over the future of open AI. Chinese labs like DeepSeek, Alibaba, and Zhipu have dominated the open-weight landscape with cheap, customizable models. Now the West is punching back: on Monday, the Nvidia-backed American startup Reflection AI unveiled its 501-billion-parameter model Beam, aimed at coding and agent tasks and benchmarked against China's GLM 5.2 and Qwen 3.8-Max. Mistral's Mistral Large 4 is Europe's counterweight in the same battle — and unlike Beam, which is still in final evaluation, Mistral has committed to a firm public release date of October 27.

For Gen Z developers, students, and builders, open-weight models are the practical on-ramp to AI: no API bills, no usage limits, full control of the data. That is why the release format matters as much as the benchmarks. If Mistral Large 4 really does close the gap with frontier models while staying downloadable, it becomes a serious option for anyone building outside the American cloud ecosystem — especially in Europe, the Gulf, and APAC, where Mensch says there is "an enormous desire and an enormous need for alternative technology suppliers."

The money is certainly there. Mistral raised €3 billion in its latest funding round in September at a valuation of €21 billion, positioning itself as a safer, European alternative to U.S. and Chinese incumbents. Mensch has publicly argued that when American executives warn AI could pose an existential threat to humanity, their comments are self-serving — a jab at rivals who simultaneously raise the alarm and race ahead. Now Mistral has to prove the technology matches the talk.

For now, the benchmark claims deserve healthy skepticism: Mistral said Large 4 beats Chinese models on cyber tasks without naming the models or the tests. Independent testing will settle that after October 27, when researchers can actually download and evaluate the weights themselves. But the strategic move is clear — Europe is done watching the AI race from the sidelines, and it just announced its biggest bet yet with Mistral Large 4. Read the Reuters report on Mistral's announcement for the full details from Abu Dhabi.