Mistral Large 4 (ML4), nicknamed Le Chonk, is a natively multimodal model with a 1-trillion-parameter latent and 49 billion active parameters, released as a public preview with weights scheduled at month’s end. Trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral’s European datacenters, it’s positioned as an open-weight, sovereign AI that can run on private cloud or on-premise. The preview is being red-teamed with cybersecurity partners and state authorities; the model emphasizes enterprise control, multilingual coverage (training data spanning 160+ languages including all EU official languages), and availability across multiple regions with a fully European deployment option.
Substantively, ML4 is presented as state-of-the-art among open models across cybersecurity, coding, agentic workflows, multimodal vision, science/math, and knowledge work. Key metrics: on cybersecurity it ranks among the global top five on the Artificial Analysis Cyber Index, scores 82% on a vulnerability reproduce-and-patch test, and solves 93% of Cybench challenges. Coding scores include DeepSWE 61.7%, Terminal-Bench 28.3%, and a Coding Agent Index of 49.8%; a blind human coding eval rated it 3.74 (second of five). Agentic and knowledge-work benchmarks include AutomationBench 59.9% and AA-Briefcase 1,393 Elo. Visual grounding exceeds some closed models (Dense200 42% vs 41% for GPT-6-Astra). Scientific and math workflows perform strongly (SciCode-Verified), and safety/robustness show high resistance to prompt-injection attacks (Lakera B3 93.3%, KORA 1.691).
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
Research shows that language models route questions about malicious code through pathways associated with morality, not just technical analysis. When asked if code is malicious, models consider moral implications, influenced by how they process related questions. (manifold.security)
Google AI Studio announced Nano Banana 2.1, an improved image generation model with better visual design, mask-based editing, and natural-looking images. The model outperforms previous versions across all metrics and is available for testing at ai.studio. (twitter.com)
AI has developed the ability to create its own inference hardware, exemplified by the open-source OpenTPU project. This advancement enables AI systems to design dedicated hardware for improved performance and efficiency. (github.com)
EmbeddingGemma 2 is a new multimodal embedding model developed by Google, capable of handling both text and vision data. It offers a moderate size of 270 million parameters for text and 440 million for combined text and vision, aiming to improve how large language models and AI agents work. (blog.google)
Etched, an AI chip startup, has received investment offers exceeding $40 billion, more than doubling its recent valuation. The company designs proprietary hardware that promises faster, cheaper inference, attracting major investors and customers like Jane Street. (techcrunch.com)
Codemode introduces a way to integrate tools directly into the execution environment for language models, allowing more complex and native interactions. It emphasizes the separation between the trusted harness and the target environment where tools run, enabling better security and functionality. (lucumr.pocoo.org)
Today's best Hacker News stories, summarized and screenshotted, one email a day.