Open Models
Open-weight and open-source language models. The foundations that power the open AI ecosystem.
The open model landscape
Open models are the foundation of the entire open source AI ecosystem. Without capable, accessible models, nothing else works — no local AI, no open agents, no independent AI infrastructure.
The good news: open models have never been better. The gap between open and closed models continues to narrow, and for many practical tasks, open models are now the right choice.
Leading model families
Llama (Meta)
The model family that catalyzed the open AI movement. Llama 3 and 4 series offer competitive performance across reasoning, coding, and general tasks.
Qwen (Alibaba)
Strong multilingual models with excellent coding performance. Qwen 2.5 and 3 series are top-tier for their size classes.
Mistral
European AI lab producing efficient, capable models. Known for strong performance-per-parameter ratios.
DeepSeek
Research-focused models with strong reasoning capabilities. DeepSeek-V3 and R1 pushed boundaries in open model performance.
Gemma (Google)
Google’s open model family, optimized for efficiency and on-device deployment.
What to watch
- Reasoning-specialized open models closing the gap with o1/o3
- Mixture-of-experts architectures enabling larger models at lower compute
- Fine-tuning and adaptation tooling becoming more accessible
- Open model licensing evolving (truly open vs. restricted open)