GLM-5-Turbo

GLM-5-Turbo, released in March 2026, is the odd member of the GLM-5 family. Where GLM-5 and GLM-5.1 are general foundation models shipped with open weights, this one is tuned from the training stage onwards for a single target: agent workloads running in the OpenClaw environment. Z.ai's own documentation calls it an "OpenClaw native model" and lists what it optimised for — tool invocation that does not fail mid-chain, decomposition of long multi-layered instructions, scheduled and long-running tasks that must survive across time, and high-throughput execution on long logical chains. The published specification is deliberately narrow: text in, text out, a 200,000-token context window, up to 128,000 tokens of output, thinking modes, streaming, function calling, context caching, structured JSON output and MCP tools. Pricing sits between the two flagships at $1.20 per million input tokens and $4.00 per million output, with cached input at $0.24. Two things are worth noting for readers comparing this with the rest of the family. First, Z.ai published a purpose-built benchmark alongside it — ZClawBench, an end-to-end test for agent tasks in that same ecosystem — and reports in the accompanying material that use of modular "skills" among such agents rose from 26% to 45% in a short period, which is the trend the model is chasing. A benchmark authored by the vendor of the model it measures should be read with that in mind. Second, unlike GLM-5, GLM-5.1 and GLM-5.2, no weights for GLM-5-Turbo have been published on Hugging Face: this is an API-only model, so the licence question that makes the rest of the family notable simply does not apply here.

▸Videos

No videos yet.

▸Reviews