Models & AIPractitioner
Open Source LLMs
Frontier-quality AI models anyone can download, run, and modify.
Open source LLMs are models whose weights are publicly released — you download them, run them on your own hardware or a cloud VM, and pay no API fees. Quality has converged rapidly with closed models.
Best models by use case (mid-2026):
- General chat & reasoning: Llama 4 Maverick (Meta, MoE, 1M ctx), Qwen3-235B (Alibaba)
- Coding & agentic: DeepSeek V4 (1.6T params, rivals GPT-5.5), Devstral Small 24B (Mistral), Qwen3-Coder
- Compact / edge: Llama 4 Scout (17B active, 10M ctx), Gemma 4 E4B (Google, runs on phones), Mistral Small 4
- Long context: Llama 4 Scout (10M tokens), Mistral Medium 3.5 (256K, open-weight)
- Multilingual: Qwen3-7B (29 languages), Gemma 4 (140+ languages)
Run via Ollama, LM Studio, or cloud APIs (Together AI, Groq, Fireworks AI). Apache-2.0 models allow commercial use without restriction.
In plain terms
Like open-source software (Linux, Firefox): the community can inspect, improve, and redistribute the model — no vendor lock-in, no usage caps.