glm-5.3-flash:cloud

Cloud tags

A single cloud tag serves the 320B MoE model (18B active) on Ollama's hosted infrastructure in the United States and Europe, with zero data retention per Ollama's privacy policy.

Benchmark highlights — GLM-5.3-Flash vs. GLM-5.2

Key features

• Agentic coding — built for sustained, multi-step coding in agent harnesses.
• Long context — a 1M context window holds project-scale codebases and long-running sessions.
• Flexible thinking — reasoning is always on, with effort tunable per request across low, high, and max levels.
• Multimodal input — accepts image and video alongside text, for workflows that mix code with visual context.
• License — MIT.

Best practices

• Thinking effort — reasoning is always on; tune effort per request across low, high, and max levels to balance quality and cost.
• Vision in the coding loop — the model can inspect its own rendered output (interfaces, games, 3D) and refine it iteratively; it also reads documents, spreadsheets, and dashboards directly as images.
• Long context — the 1M window plus sparse attention (4.4× smaller KV cache) keeps long sessions and large codebases efficient.
• Efficient inference — linear + sparse attention with IndexPool cuts attention compute 3.0× vs. GLM-5.3, making the flash tier cheap at scale.