4 ms·

They implement more openai-compatible endpoints than ollama at least