AI PLATFORMS & MODELS
OPEN WEIGHTSOPEN SOURCE SELF-HOSTABLE
Mistral Small 3
by Mistral AI
Low-latency chat assistants and lightweight agentic tasks
85HYLAND SCORE
Overview
Mistral Small 3 is a 24B-parameter, latency-optimized language model released under Apache 2.0 that competes with much larger models on instruction-following and reasoning benchmarks. It runs efficiently on a single GPU, making it a strong choice for self-hosted assistants, agents, and fine-tuning workflows.
Key features
- Best for: Low-latency chat assistants and lightweight agentic tasks
- 32k tokens context window
- API access available
- Open weights — self-host and fine-tune
- Free tier: Free tier on La Plateforme; weights free to download
Tags
APISelf-hostApache 2.0
Related in AI Platforms & Models
FEATURED VERIFIED
DeepSeek R1
DeepSeek
93HYLAND
Open-weight reasoning that rivals the frontier.
Open WeightsOpen Source Self-host
15.3k/wk
VERIFIED
Llama 3.1 70B
Meta
92HYLAND
Self-hosted AI, privacy-critical apps
Open WeightsOpen Source Self-host