1. SovInfra

    French AI inference service on European GPUs: open-weight models behind an OpenAI-compatible API, no data retention, $0.12 per million input tokens and 1 billion free tokens.