Mistral Small 3.1: How a 2.4B‑Parameter Open‑Source Model Challenges GPT‑4

OpenAI's competitor Mistral has open‑sourced a lightweight AI model.

Paris‑based AI startup Mistral AI reports the release of Mistral Small 3.1 , a new lightweight model that uses only 2.4 billion parameters yet claims performance superior to comparable models from OpenAI and Google.

The model handles both text and images, offers an expanded context window of up to 128,000 tokens, and can process data at roughly 150 tokens per second, making it suitable for fast‑response applications.

More Convenient, More Sustainable

The strategy focuses on accessibility: the model runs efficiently on modest hardware such as a single RTX 4090 GPU or a laptop with 32 GB RAM, enabling advanced AI applications on smaller devices and remote locations.

This approach may prove more sustainable than continuously scaling model size. As companies like DeepSeek adopt similar tactics, better‑funded competitors may eventually need to follow suit.

Mistral AI was founded in 2023 by former researchers from Google DeepMind and Meta, quickly becoming a leading European AI firm. To date it has raised over $1.04 billion, valuing the company at roughly $6 billion—substantially less than OpenAI's $80 billion valuation.

The Power of Open Source

One of Mistral's distinctive choices is to keep its models open source, contrasting with the closed, proprietary approaches of many rivals. This has spurred the development of high‑quality inference models based on Mistral Small 3 and demonstrates how collaborative openness can accelerate AI progress.

While open‑source models benefit from broader community contributions, they also pose revenue challenges, prompting Mistral to focus on professional services, enterprise implementations, and unique AI applications.