Open-Weight Models

All five engines say yes, but with a catch. ChatGPT, Gemini, Google AI Mode and Perplexity say true open source means more than downloadable weights, and Perplexity names OLMo, Pythia, Amber and CrystalCoder as meeting that bar. Copilot is the loosest, counting frameworks and models such as PyTorch, Llama and Mistral.

All five engines agree that open-weight models release downloadable weights that you can run and fine-tune yourself, while closed-weight models are reachable only through an API or app. They cite Llama, DeepSeek and Mistral as open examples and OpenAI, Anthropic and Google as closed ones, with no dissent.

All four engines back the use of open-weight models and none dissents, though they list different families, with Mistral named by ChatGPT and Copilot.

All four engines say companies pay for reliability, security and lower operating burden rather than raw intelligence. Copilot alone says proprietary models still lead on accuracy and latency.