Hetzner launches experimental OpenAI-compatible API for Qwen2.5-35B-A3B-FP8
July 24, 2026
Hetzner is testing an OpenAI-compatible inference API running on its own hardware. The current experimental offering features the Qwen/Qwen2.5-35B-A3B-FP8 MoE model, which supports text and image inputs with a 262K context window.
HOW THIS AFFECTS YOU
●
builderYou can test low-cost inference using your existing OpenAI client logic.
●
founderThis provides a potential alternative to hyperscalers for hosting workloads on specialized infrastructure.