Mistral AI announced the release of Mistral Small 4 on March 16, 2026, introducing a new model that integrates the capabilities of its flagship models into a single, versatile solution. The model unifies the strengths of Magistral for reasoning, Pixtral for multimodal tasks, and Devstral for coding agents, offering users a unified experience without needing to switch between specialized models. Mistral Small 4 is released under the Apache 2.0 license, continuing the company's commitment to open-source AI. The model is designed for general chat, coding, and complex reasoning tasks, with support for both text and image inputs. Mistral AI emphasized its role in the NVIDIA Nemotron Coalition as a founding member, highlighting its commitment to collaborative AI innovation. Key architectural features include a Mixture of Experts (MoE) system with 128 experts, 4 active per token, and 119B total parameters. The model supports a 256k context window, allowing for long-form interactions. Users can configure reasoning effort, choosing between fast, low-latency responses and deep, reasoning-intensive outputs. Performance improvements include a 40% reduction in end-to-end completion time in latency-optimized setups and 3x more requests per second in throughput-optimized configurations compared to Mistral Small 3. Benchmarks show Mistral Small 4 with reasoning achieves competitive scores, matching or surpassing GPT-OSS 120B on multiple tests, while generating significantly shorter outputs. On AA LCR, Mistral Small 4 scores 0.72 with 1.6K characters, while Qwen models require 3.5-4x more output for similar performance. On LiveCodeBench, Mistral Small 4 outperforms GPT-OSS 120B with 20% less output. Mistral Small 4 is available through the Mistral API, AI Studio, and Hugging Face Repository. Developers can prototype it for free on NVIDIA accelerated computing platforms, and for production use, it is available as an NVIDIA NIM. The model can also be customized with NVIDIA NeMo for domain-specific fine-tuning. *Source: [mistral](https://mistral.ai/news/mistral-small-4/)*