Mistral Small 4 TL5
Mistral Small 4 is Mistral AI's open-weight multimodal model combining everyday instruction following, deep reasoning, and agentic coding in one system. Its Mixture-of-Experts architecture contains 119B parameters across 128 experts, selecting four per token and activating 6.5B parameters. With a 256K-token context window, it accepts text and images and generates text for document parsing, visual analysis, codebase exploration, research, and enterprise assistants. Developers can switch reasoning effort between none for fast responses and high for complex tasks. Native function calling and structured JSON output support coding agents and automated workflows. Optimized inference options include an Eagle speculative-decoding head and NVFP4 weights. The model is released under Apache 2.0 for commercial use.
Mistral Small 4 TL5
Mistral Small 4 is Mistral AI's open-weight multimodal model combining everyday instruction following, deep reasoning, and agentic coding in one system. Its Mixture-of-Experts architecture contains 119B parameters across 128 experts, selecting four per token and activating 6.5B parameters. With a 256K-token context window, it accepts text and images and generates text for document parsing, visual analysis, codebase exploration, research, and enterprise assistants. Developers can switch reasoning effort between none for fast responses and high for complex tasks. Native function calling and structured JSON output support coding agents and automated workflows. Optimized inference options include an Eagle speculative-decoding head and NVFP4 weights. The model is released under Apache 2.0 for commercial use.