Hermes Agent users can now integrate a new high-speed inference model through external API providers. The GLM-5.3 FlashX from Zhipu AI is available via the Nous Portal and OpenRouter platforms, offering a generation speed that is five times faster than the previous Flash version. This iteration delivers peak output of 200 tokens per second.

Zhipu AI charges ¥2 per million input tokens and ¥7 per million output tokens, a 2.5 times price increase over the original GLM-5.3-Flash. The model runs on infrastructure supported by approximately 100,000 Chinese AI chips, which were optimized by a GLM-5.3 powered agent in less than two weeks to triple end-to-end throughput.