Mistral Small but Mighty - Apache 2.0, Multimodal & Fast

Introducing mral Small 3.1, a state-of-the-art multimodel multilingual model released under Apache 2.0 by Google. The model boasts an expanded context window of 128 and delivers impressive performance in terms of tokens per second compared to smaller models.

mral Small 3.1 outperforms or is comparable to smaller models, demonstrating strong performance across different benchmarks. It consistently outperforms other models in its category and excels in both multimodal and multilingual benchmarks.

A comparison of mral Small 3.1 with Gemma 3 and gbd4 or mini models in terms of performance on benchmarks like chart and multilingual capabilities. The model shows strong performance but lags behind in some specific tests.

Exploration of how the system prompt interacts with users, providing responses based on the context and ensuring accuracy in information retrieval. The prompt handles requests effectively and avoids providing inaccurate information.

Demonstration of mral Small 3.1's capabilities in understanding images and providing accurate responses based on image content. The model can analyze images, read text, and provide detailed information about visual content.

Examples of using mral Small 3.1 API for text and image processing, including classification tasks like choosing the best French cheese and identifying email content. The model demonstrates proficiency in processing text and images effectively.

Capabilities of mral Small 3.1 in analyzing data and providing insights based on visual charts. The model can understand data trends, percentages, and provide detailed analysis based on the data presented.

Demonstration of the model's ability to transcribe text accurately and convert plain text into structured formats like JSON. The model's transcription capabilities are showcased with examples of text transcription tasks.