Mistral's Ministral 3 3B Instruct 2512 is a compact, open-source language model with vision capabilities, fine-tuned for instruction tasks and ideal for chat and instruction-based use cases. It excels at tasks such as image captioning, text classification, and real-time efficient translation, making it suitable for lightweight, real-time applications on edge or low-resource devices.
Input
Output
Context
262K
Max Output
131K
Parameters
4.3B
Input Modalities
Output Modalities
Estimates based on INT8 quantization. Actual requirements vary by framework and configuration.
Data sourced from official provider APIs and documentation
Last updated: Jun 23, 2026
Automatically route workloads to the right model for every task, every time.