Run GLM-5.3-Flash in the EU through an OpenAI-compatible API

GLM-5.3-Flash was released by Z.ai in late August 2026. Unlike GLM-5.3, it starts from a newly trained base model, pre-trained on a 30-trillion-token multimodal corpus. It is a mixture-of-experts model with 320 billion parameters in total and 18 billion active per token, with 45 layers, roughly half the depth of earlier large GLM models.

It is the first GLM-5 model to accept images natively. For the first time in the GLM series it combines sparse and linear attention, which lowers the cost of very long contexts, and it uses manifold-constrained hyper-connections to improve training efficiency. It reasons before answering, with effort levels low, high and max, and its weights are MIT-licensed.

GLM-5.3-Flash is designed to give much of the flagship's coding and agent ability at a fraction of the compute. Z.ai reports that it outperforms GLM-5.2 across benchmarks and real workloads. It suits coding assistants, agents that read screenshots, and long-document work where the one-million-token window matters but a 750-billion-parameter model would be excessive.

On the hardest long-horizon tasks, the full GLM-5.3 or Kimi K3 still go further. Its reasoning defaults to the maximum level, which produces long outputs unless you lower it. The model card lists English and Chinese as its primary languages, so test European languages on your own content. It reads images but does not create them.

inference.eu speaks both the OpenAI and the Anthropic request formats. Set the base URL to the inference.eu API, use your inference.eu key and set the model to glm-5.3-flash. Coding tools such as Claude Code, Cline or OpenHands work by changing those same three settings.

Send images as image_url content parts next to your text. Pass reasoning_effort as low, high or max; Z.ai defaults to max, so set low explicitly for quick replies. For chat, Z.ai recommends clearing earlier thinking from the history. Tool calls, JSON output and streaming work as usual.

On inference.eu, GLM-5.3-Flash runs on infrastructure in the European Union. Your prompts, images, source code and the model's responses are processed in EU data centres, are not sent to Z.ai in China, and are never used to train any model.

Every customer receives a data processing agreement under Article 28 of the GDPR together with our subprocessor list, so you can document every party involved. Billing is in euros. Unlike the GLM-5.3 flagship, GLM-5.3-Flash is released under the plain MIT licence, which allows commercial use with no revenue thresholds or review requirements.