Ministral 3 14B (Infomaniak)

Mistral AI released Ministral 3 14B in December 2025 as part of the Ministral 3 family, with the instruct variant published on Hugging Face under the identifier Ministral-3-14B-Instruct-2512. The model is presented by third-party directories as a dense transformer built around a 14-billion parameter language core, intended to deliver high-performance intelligence in a footprint small enough for edge devices and single-GPU setups. Its design targets low-latency inference on resource-constrained hardware while preserving capabilities that approach those of larger frontier models, making it well suited to private or on-premises deployments where data locality matters.

A large 256,000-token context window lets the model work through long documents and multi-step agentic workflows, and it ships with features aimed at autonomous tool use such as native function calling and structured JSON output. Third-party benchmark aggregators place Ministral 3 14B in the middle tier of contemporary open models, reflecting a balanced profile rather than a specialist focus on any single task. Practically, it fits teams that want an open-weights Mistral model for code assistance, document analysis, and integration into pipelines that require long context and reliable function calling without the cost of larger hosted alternatives.