Z.ai unveils a proprietary GLM-5-Turbo model for agent workflows, signalling a shift from open source distribution to monetised, closed AI systems for enterprise use.
Z.ai has introduced GLM-5-Turbo, a faster and slightly cheaper proprietary variant of its open-source GLM-5 model, signalling a notable shift in its open-source strategy. Designed for agent-based workflows, the model targets OpenClaw-style use cases including tool use, long-chain execution and persistent automation.
Unlike the flagship GLM-5, which is open-source under an MIT licence, GLM-5-Turbo is closed-source and positioned as a commercial, execution-focused offering. Z.ai states that while innovations from Turbo will inform future open source releases, there is no commitment to open-source the model itself.
The model prioritises execution over conversation, with optimisations for complex instruction decomposition, structured tool use and multi-step task completion. It is aimed at enterprise AI agents handling automation, coding, workflow orchestration and data analysis.
GLM-5-Turbo offers a ~202.8K token context window and 131.1K max output, with support for tool calling and structured outputs. It delivers 48 tokens per second throughput and an 8.16-second completion time, alongside a significantly lower 0.67% tool error rate. Pricing is set at $0.96 per million input tokens and $3.20 per million output tokens, marginally cheaper than GLM-5.
Available via API on OpenRouter and through Z.ai’s GLM Coding subscriptions, the release underscores a broader shift. Open-source models continue to drive adoption, while proprietary variants capture enterprise value—an approach increasingly mirrored across the global AI landscape.
