Mistral Small 4 Released: First to Combine Reasoning, Multi-modal, and Programming Capabilities

Mistral AI, a leading European AI lab, recently launched the Mistral Small 4 model. It is the lab's first comprehensive large language model (LLM) integrating the strengths of flagship reasoning, multi-modal, and programming models, allowing users to access full functionality without switching between different models. The model uses a MoE architecture with 119B total parameters and 6B active parameters, a 256k context window, supports fast and deep reasoning modes, and is open-sourced under the Apache 2.0 license. Compared to its predecessor Small 3, it reduces completion time by 40% in latency-optimized mode, triples request volume in throughput-optimized mode, and performs no worse than OpenAI's GPT-OSS 120B in benchmark tests.