Mistral Small 4
7 hours ago
- #AI
- #OpenSource
- #Multimodal
- Mistral Small 4 is announced, unifying Magistral (reasoning), Pixtral (multimodal), and Devstral (coding) into one versatile model.
- Released under Apache 2.0 license, supporting open, accessible, and customizable AI.
- Hybrid model optimized for chat, coding, agentic tasks, and complex reasoning with text and image inputs.
- Key features: Mixture of Experts (MoE), 119B total parameters, 256k context window, configurable reasoning effort, and native multimodality.
- Performance highlights: 40% reduction in latency, 3x more requests per second compared to Mistral Small 3.
- Enterprise-grade efficiency with optimized setups for NVIDIA hardware.
- Competitive benchmarks: Matches or surpasses GPT-OSS 120B with shorter outputs.
- Designed for developers, enterprises, and researchers with open-source flexibility.
- Available via Mistral API, AI Studio, and NVIDIA NIM for optimized deployment.