Mistral The small model of Artificial Intelligence 3.1 (AI) was launched on Monday. The Paris-based AI company introduced two open source variants of the latest model-blast model and instructions. The model is the successor of Mistral Small 3 and offers better text performance and multimodal understanding. The company says it exceeds comparable models such as Google Gemma 3 and the OpenAi GPT-4O Mini on multiple benchmarks. One of the main advantages of the newly produced model is its quick response time.
Mistral Small 3.1 Model Ai Launched
In a newsroom publishThe AI company detailed the new models. Mistral Small 3.1 comes with an expanded context window of up to 1,28,000 tokens and is said to provide inference speeds of 150 tokens per second. This essentially means that the AI model response time is quite fast. It arrives in two chat variants and instructions. The first acts as a typical chatbot, while the latter is set to follow the user instructions and is useful when creating an application with a specific goal.
Mistral Small 3.1 Benchmark
Photo credit: Mistral
Similar to previous releases, Mistral Small 3.1 is available in the public domain. Open weights can be downloaded from the face embracing the company Listing. The AI model comes with an Apache 2.0 license that allows academic and research use, but prohibits cases of commercial use.
Mistral said the Large Language Model (LLM) is optimized to run on a single NVidia RTX 4090 GPU or a 32 GB RAM device. This means that enthusiasts without an expensive configuration to run AI models can also download and access it. The model also offers low latency function calls and functions, which can be useful for construction automation and agentic workflows. The company also allows developers to adjust Mistral Small 3.1 to adjust the cases of use of specialized domains.
Reaching the performance, the IA company shared several reference scores based on internal testing. Mistral Small 3.1 is said to have surpassed GEMMA 3 and GPT-4O Mini on Postgraduate Test questions and Answers on Google (GPQA) Main and Diamond, Humaneval, Mathvista and the dovqa benchmarks. However, the GPT-4O Mini performed better in the huge multitasking (MMLU) languages, and Gemma 3 exceeded it in mathematical benchmark.
In addition to hugging the face, the new model is also available through the application programming interface (API) at AI AI developer, La Plackform, as well as the AI Vertex of Google Cloud. It will also be made available at Nvidia’s NIM and Azure Ai Foundry of Microsoft in the coming weeks.