AI Pulse · AI Trends Pulse

Zhipu has released GLM-5.3-Flash, an open, multimodal model aimed at coding and AI agent work. It uses a mixture-of-experts design with 320 billion total parameters, though only 18 billion are active for a given task. That matters because it can offer the reach of a much larger model without paying the full compute cost every time. It also has a 1 million token context window, an MIT licence, and pricing of $0.15 and $0.50 per million tokens.

Z.ai says the model outperforms its earlier GLM-5.2 on coding and agent-style tasks at one tenth of the price. It also says GLM-5.3-Flash comes close to Claude Opus 4.8 on Z.ai’s own coding benchmark. Treat that second claim with care. It is the vendor’s benchmark, not an independent comparison. Still, the direction is clear, and the details are available in the original discussion.

For a business owner, this is another reminder that model selection is now a cost decision as much as a capability decision. The most expensive model may still be right for high-stakes work, but routine coding, document processing, support drafting, and internal agents may not need it. Teams should test two or three models against their actual tasks, measure quality, speed, and cost, then route work accordingly. This is the kind of thing we build into an AI command centre, so teams can see which model is doing what and what it is costing.