Open-source AI,
no infrastructure.

Meta's flagship open-weight models. Llama 3.3 70B matches GPT-4o on many benchmarks at a fraction of the cost.

Chinese open-source lab producing state-of-the-art coding and reasoning models. DeepSeek R1 rivals o1 at open-weight pricing.

European open-source AI models known for efficiency. Mistral 7B and Mixtral 8x7B set the standard for small, capable models.

Qwen2.5 and QwQ are among the strongest open-weight models for coding and reasoning tasks, including 128k context variants.

Google's open-weight Gemma models are compact and highly capable — Gemma 3 27B competes with models twice its size.

Microsoft's Phi models punch far above their weight class. Phi-4 14B performs comparably to much larger models on reasoning tasks.