Mistral AI Unveils Mistral Small 4: A Versatile Model with Reasoning and Multimodality

Mistral AI announced Mistral Small 4, the latest addition to the Mistral Small family. This is the first Mistral model to combine the capabilities of its flagship models—Magistral for reasoning, Pixtral for multimodal, and Devstral for agentic coding—into a single versatile model. It uses a Mixture of Experts architecture with 128 experts, 4 active per token, totaling 119B parameters but only 6B active per token. The model supports a 256k context window, accepts both text and image inputs, and features a configurable reasoning effort parameter. Mistral Small 4 is released under the Apache 2.0 license, making it fully open source. Performance highlights include a 40% reduction in end-to-end completion time and up to 3x more requests per second compared to Mistral Small 3. The model is available via Mistral API, AI Studio, and as an NVIDIA NIM, with support for customization using NVIDIA NeMo. Mistral AI has also joined the NVIDIA Nemotron Coalition as a founding member.