Mistral Small 4 119B is a hybrid model from Mistral AI designed to behave like three specialists in one. It folds the lineage of the Instruct family, the reasoning-focused Magistral track, and the Devstral coding line into a single system, with the ability to switch between a fast instant-reply mode and a deeper reasoning mode on a per-request basis. The architecture uses a mixture-of-experts design with 128 experts and 4 active per token, totaling 119B parameters while activating only about 6.5B per forward pass, which is what allows such a large model to remain economical at serving time. Multimodal input for both text and image combined with text output makes it equally comfortable in conversational, analytical, and visual-understanding settings, and a very long context window positions it for document-heavy and code-repository workloads.

The practical appeal of Mistral Small 4 lies in how that architectural lineage translates into day-to-day strengths. Compared with Mistral Small the listed price, latency-optimized deployments see a 40 percent reduction in end-to-end completion time, and throughput-optimized deployments handle roughly three times more requests per second, with additional efficiency unlocked through a trained EAGLE speculative-decoding head and a 4-bit NVFP4 quantized checkpoint. The reasoning effort can be configured per request, so the same model can power a snappy customer-support assistant one moment and a deep legal or financial research agent the next, while function calling, tool orchestration, and vision capabilities let it serve as the backbone for agentic and enterprise copilot pipelines. Together these traits make it a flexible general-purpose model aimed at teams that want strong reasoning and multimodal understanding without paying the full inference cost of a dense flagship.