Mistral Small (latest)

Mistral Small is the compact tier in Mistral AI's lineup, positioned for developers who want a small, fast model that can run on local hardware while still covering a useful slice of general-purpose tasks. The family lineage is anchored by Mistral Small 3, which Mistral AI described as a latency-optimized 24-billion-parameter model released under the Apache 2.0 license, signaling a shift from earlier Mistral Research License terms toward permissive open weights for general-purpose releases.

In practical terms, Mistral Small is aimed at workloads where quick responses and the ability to self-host matter more than top-end reasoning, such as chat assistants, on-device copilots, retrieval-augmented agents, and prototyping against larger proprietary systems. Mistral AI positioned Small 3 as competitive with much larger models like Llama 3.3 70B and Qwen 32B while running more than three times faster on the same hardware, and offered it as an open replacement for opaque smaller proprietary models, a framing that continues to shape how the Small line fits into mixed-model deployments today.