Mistral Large 4 is a high-capacity, open-weight multimodal model built with a granular Mixture-of-Experts architecture, offered in public preview (Oct 6, 2026). It exposes 52 billion active parameters and a total of 1.05 trillion parameters, plus a 1.6 billion-parameter vision encoder, and supports an extremely long 1 million-token context window. The release is positioned as a general-purpose foundation for multimodal tasks, combining language and vision capabilities and a labeled variant (mistral-large-4+1) aimed at balancing speed and performance.
Practical deployment details emphasize API-first use: per‑million token pricing is shown (input ≈ $1.36, cached input ≈ $0.14, output ≈ $4.18), a web playground is available, and the model supports common production features such as structured outputs, function calling, document Q&A, prefixing, chat completions, batching, agents and conversations, and built‑in tools. The product listing highlights endpoints for chat completions, batch inference, and agents, indicating readiness for integrated applications that require multimodal understanding, long-context reasoning, and scalable inference.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.