Mistral Small 3.1 AI Model With Improved Text and Multimodal Performance Released

Mistral Small 3.1 synthetic intelligence (AI) mannequin was launched on Monday. The Paris-based AI agency launched two open-source variants of the most recent mannequin — chat and instruct. The mannequin comes because the successor to the Mistral Small 3, and provides improved textual content efficiency and multimodal understanding. The firm claims that it outperforms comparable fashions akin to Google’s Gemma 3 and OpenAI’s GPT-4o mini on a number of benchmarks. One of the important thing benefits of the newly launched mannequin is its speedy response instances.

Mistral Small 3.1 AI Model Released

In a newsroom submit, the AI agency detailed the brand new fashions. The Mistral Small 3.1 comes with an expanded context window of as much as 1,28,000 tokens and is claimed to ship inference speeds of 150 tokens per second. This basically means the response time of the AI mannequin is kind of quick. It arrives in two variants of chat and instruct. The former works as a typical chatbot whereas the latter is fine-tuned to observe consumer directions and is beneficial when constructing an utility with a selected function.

Mistral Small 3.1 benchmark
Photo Credit: Mistral

Similar to its earlier releases, the Mistral Small 3.1 is obtainable within the public area. The open weights could be downloaded from the agency’s Hugging Face itemizing. The AI mannequin comes with an Apache 2.0 licence which permits educational and analysis utilization however forbids industrial use instances.

Mistral stated that the big language mannequin (LLM) is optimised to run on a single Nvidia RTX 4090 GPU or a Mac gadget with 32GB RAM. This means lovers with out an costly setup to run AI fashions may obtain and entry it. The mannequin additionally provides low-latency perform calling and perform execution which could be helpful for constructing automation and agentic workflows. The firm additionally permits builders to fine-tune the Mistral Small 3.1 to suit the use instances of specialized domains.

Coming to efficiency, the AI agency shared varied benchmark scores primarily based on inside testing. The Mistral Small 3.1 is claimed to outperform Gemma 3 and GPT-4o mini on the Graduate-Level Google-Proof Q&A (GPQA) Main and Diamond, HumanEval, MathVista, and the DocVQA benchmarks. However, GPT-4o mini carried out higher on the Massive Multitask Language Understanding (MMLU) benchmark, and Gemma 3 outperformed it on the MATH benchmark.

Apart from Hugging Face, the brand new mannequin can be accessible by way of the appliance programming interface (API) on Mistral AI’s developer playground La Plateforme, in addition to on Google Cloud’s Vertex AI. It can even be made accessible on Nvidia’s NIM and Microsoft’s Azure AI Foundry within the coming weeks.