Source URL: https://simonwillison.net/2025/Sep/19/magistral/
Source: Simon Willison’s Weblog
Title: Magistral 1.2
Feedly Summary: Mistral quietly released two new models yesterday: Magistral Small 1.2 (Apache 2.0, 96.1 GB on Hugging Face) and Magistral Medium 1.2 (not open weights same as Mistral’s other “medium" models.)
Despite being described as "minor updates" to the Magistral 1.1 models these have one very notable improvement:
Multimodality: Now equipped with a vision encoder, these models handle both text and images seamlessly.
Magistral is Mistral’s reasoning model, so we now have a new reasoning vision LLM.
The other features from the tiny announcement on Twitter:
Performance Boost: 15% improvements on math and coding benchmarks such as AIME 24/25 and LiveCodeBench v5/v6.
Smarter Tool Use: Better tool usage with web search, code interpreter, and image generation.
Better Tone & Persona: Responses are clearer, more natural, and better formatted for you.
Tags: vision-llms, llm-release, mistral, generative-ai, llm-reasoning, ai, llms
AI Summary and Description: Yes
Summary: Mistral has introduced two new models, Magistral Small 1.2 and Magistral Medium 1.2, enhancing their capabilities through multimodality, which integrates text and image processing. These updates include significant performance improvements and enhanced tool usage, making them relevant for professionals focused on advancements in AI, particularly in the areas of LLM and generative AI security.
Detailed Description: Mistral’s recent announcement highlights the release of two upgraded models in their Magistral series, which present notable improvements over previous versions. The key enhancements are as follows:
– **Multimodality**:
– The new models incorporate a vision encoder, allowing them to process and interpret both text and images effectively. This advancement is particularly relevant as it expands the usability of LLMs (Large Language Models) into more complex applications that require understanding visual data alongside text.
– **Reasoning Capabilities**:
– As reasoning models, the Magistral series aims to enhance cognitive tasks, making them more suitable for applications that demand deeper understanding and reasoning, such as question-answering systems that involve visual elements.
– **Performance Enhancements**:
– The models exhibit a 15% improvement in performance on notable benchmarks like AIME 24/25 and LiveCodeBench v5/v6, which indicates a significant upgrade in their functionality and reliability for coding and mathematical tasks.
– **Improved Tool Utilization**:
– The new models demonstrate better integration with tools such as web search, code interpreters, and image generation applications, enhancing their versatility for developers and businesses looking for AI solutions that can handle diverse input types.
– **Refined User Experience**:
– The models provide clearer and more natural responses, which is vital for ensuring user satisfaction in interactive applications and enhances overall communication quality.
These updates from Mistral underline a trend towards more advanced and integrated AI systems, which security and compliance professionals should monitor, particularly regarding the implications of multimodal capabilities and any potential security vulnerabilities that may arise from their deployment in various environments. Keeping up with advancements in LLMs also prompts a reassessment of security measures and compliance frameworks to adequately protect integrated systems handling diverse data types.