Open-Source Models

Open-source AI models are models whose trained weights (and often training code and data details) are publicly available for download and use. Unlike proprietary models from OpenAI or Anthropic that are only accessible through APIs, open-source models can be run on your own infrastructure, giving you full control over data privacy, customization, and deployment. The open-source AI ecosystem has grown rapidly, with models now competitive with proprietary alternatives on many tasks.

Major open-source model families include Meta's Llama (8B to 405B parameters), Mistral and Mixtral from Mistral AI, DeepSeek's coding and reasoning models, Google's Gemma, Microsoft's Phi, and Alibaba's Qwen. Licensing varies: some use truly permissive licenses (Apache 2.0), while others have restrictions on commercial use or model size. Meta's Llama license, for example, allows commercial use but requires a license agreement for applications with over 700 million monthly active users.

The advantages of open-source models are significant. Data never leaves your infrastructure, which is critical for regulated industries like healthcare and finance. You can fine-tune models on proprietary data without sharing it with a third party. Costs become predictable — you pay for compute, not per token. You avoid vendor lock-in and can switch between models freely. And the active community continuously improves and extends these models.

The tradeoffs are also real. Self-hosting requires ML infrastructure expertise, GPU hardware (or cloud GPU rental), and ongoing maintenance. The absolute frontier of capability — the very best performance on the hardest tasks — still tends to come from the largest proprietary models, though the gap has been shrinking. Many organizations adopt a hybrid approach: using proprietary APIs for the most demanding tasks and open-source models for high-volume, latency-sensitive, or privacy-critical workloads. GPTCrunch tracks open-source models alongside proprietary ones so you can make informed comparisons.