mistral

Mistral has released Mistral Large 4 ('Le Chonk'), a one-trillion-parameter open-weight model it claims is the most capable open-weight model outside of China, optimized for coding, cyberdefense, and domain-specific tasks.

Mistral releases Mistral Large 4 ('Le Chonk'), a natively multimodal 1T-parameter model with 49B active parameters, claiming best open-weights performance aggregated benchmarks and state-of-the-art results on workloads like cyber defense, manufacturing, and finance. It is developed and deployable entirely in Europe, available via API now with open weights expected end of October.

Mistral is expected to release a new model today, with CEO Arthur Mensch claiming it outperforms Chinese models in some areas, highlighting the growing competition among open-weight LLM providers.

Mistral has raised €3 billion, emphasizing the need for strong European AI competitors without compromising product quality in the name of sovereignty.

Mistral has raised €3 billion in Series D funding to advance sovereign, open-weight AI, emphasizing control over data, infrastructure, and models. The round, led by Samsung Electronics, marks the largest equity fundraising by a European technology company.

Mistral's default policy now includes training on user input and output data, with opt-out options available through various platforms, except for enterprise tier customers.

Mistral introduces Agentic Search, a multi-step retrieval loop that enhances AI system accuracy and efficiency in navigating complex documents, reducing latency and token use compared to traditional RAG methods.

Mistral is now hosting competitor Z.ai's GLM-5.2, pricing it cheaper than its own flagship Mistral Medium 3.5, sparking speculation about a strategic pivot toward compute sales and smaller specialized models.

Mistral AI announces OCR 4.1, an updated OCR service with native paragraph-level bounding box extraction, structural block labels, and block-level confidence scores, now in public preview.

Mistral has been granted a patent for a method that uses an LLM to generate code blocks encapsulating tool calls, executing them in a sandbox and pausing for client-side results when needed.

Mistral releases Shieldstral-1.0-3B, a compact safety/moderator model jokingly referenced as 'Not-Hotdog', aimed at filtering harmful content.

Mistral is gaining strategic advantage as a European open-source AI lab amid US export restrictions and safety incidents involving American proprietary models, positioning itself as a sovereign alternative. The company is reportedly raising funds at a $23 billion valuation with surging revenue.

Analyzes Mistral's strategic pivot from competing on frontier models to becoming a European Palantir, focusing on enterprise AI deployment and the application layer, with revenue growth as evidence.

Austria is launching a government AI platform powered by Mistral models and Open WebUI.

A metaphorical article suggesting that Mistral AI consistently takes an unconventional or contrarian approach in the AI industry, similar to a fish swimming against the current.

Mistral AI releases Shieldstral-1.0-3B, a compact multimodal safety classifier that adapts to natural-language safety policies at inference time, supporting text, image, and text+image moderation.

Upcoming AI model releases include Kimi K3, Deepseek V4, new Liquid and Mistral models, with rumors of GLM 5.5 in August, indicating a busy period for open-weight AI.

Mistral has released its first robotics model, which can navigate using only a single RGB camera, eliminating the need for lidar.

Mistral Devs published a tutorial on building a medical document processing workflow using Mistral OCR, Agents, and Workflows, including a human-in-the-loop step for low-confidence classifications.

Mistral launched OCR 4, but a tweet points out that they reported lower scores for Chandra 2 than its actual performance and omitted Infinity Parser (87.6%) from their olmocr comparison.