mistral-small-latest

Mistral Small 3.1 is a lightweight multimodal language model launched by Mistral AI, designed as an efficient inference engine that strikes a balance between being “small yet powerful, fast yet comprehensive,” while delivering robust capabilities in text, image, multilingual support, and long-context understanding.

• Outstanding overall performance: In multiple benchmarks—including MMLU, HumanEval, and MMMU—Mistral Small 3.1 outperforms peer models such as Gemma 3 and GPT-4o Mini.
• Native multimodal understanding: It can directly process image-plus-text inputs and supports visual tasks like chart interpretation and screenshot-based question answering.
• Ultra-long context of 128K tokens: Easily handles long-document summarization and complex multi-turn dialogues, with consistently reliable retrieval of key information.
• Deep multilingual support: Covers major language families across Europe, East Asia, the Middle East, and more, significantly improving performance in non-English scenarios.

───────────────────────────────────────────────────────────────────

Core Capabilities

⚡ High-speed, low-consumption inference: Generates up to 150 tokens per second, offering fast response times and low costs, making it ideal for high-concurrency business applications. 👁️ Integrated image-and-text analysis: Understands screenshots of interfaces with embedded text, data charts, or product images, and provides precise answers based on contextual understanding. 🌍 Natural cross-language interaction: Outputs responses that align with local linguistic conventions, avoiding mechanical translation and truly enabling global usability. 🧩 Agent-friendly architecture: Supports Function Calling and structured responses, allowing seamless integration into automated workflows or AI assistants.