Zhipu AI launches GLM-5-Turbo: a model built exclusively for OpenClaw AI agents
Chinese AI startup Zhipu AI (operating internationally under the brand Z.ai) has unveiled GLM-5-Turbo, a large language model designed from the outset for a very specific task: working in OpenClaw scenarios — that is, complex, automated agent workflows. With this, Zhipu AI is taking a different path than many competitors, who have so far adapted general-purpose models retroactively for agent tasks.
Context
GLM-5-Turbo builds on the earlier GLM-5 architecture released in February 2026. The base model itself introduced several technical upgrades over previous versions, including larger training datasets reaching 28.5 trillion tokens, 745 billion total parameters (44 billion active), and a context window of 200,000 tokens.
Zhipu AI has become China’s first public AI company to ship a frontier model. The company completed its Hong Kong IPO in January 2026, raising approximately HKD 4.35 billion (USD $558 million), and is currently valued at approximately $34.5 billion.
What is OpenClaw?
OpenClaw designates an ecosystem for AI agents in which models do not merely answer questions but independently plan tasks, invoke tools, and execute multi-step processes. The user base has expanded significantly: in addition to developers, financial professionals, operations engineers, content creators, and research analysts now use such agent systems.
Particularly striking is the rapid increase in so-called skills — modular capabilities that agents can flexibly deploy. Zhipu AI reports that the share of skills in OpenClaw workflows has risen from 26 to 45 percent.
GLM-5-Turbo Details
Competitive Pricing
GLM-5-Turbo is significantly cheaper via API than other top LLMs; it costs just $1.2 (input) or $4 (output) per million tokens. For comparison: Anthropic’s Claude Opus 4.6 is five times more expensive at $5 and $25 (input/output per 1M tokens).
Technical Specifications
- Context window: 200,000 tokens
- Maximum output: 128,000 tokens in a single response
- Architecture: Mixture of Experts (MoE) with 745B total parameters, 44B active
- Training hardware: Huawei Ascend (US-independent)
Key Capabilities
Tool Invocation: The model can invoke external tools precisely and reliably. Especially for multi-step tasks, it is said to work more stably and with fewer errors than previous models.
Instruction Understanding: GLM-5-Turbo understands complex, multi-layered instructions, breaks them down into sub-steps, and supports collaboration among multiple agents in task distribution.
Time-Triggered and Continuous Tasks: The model is specifically optimized for scenarios in which tasks are triggered at specific times or should run continuously over extended periods.
High Throughput with Long Chains: For tasks with large data volumes and long logical chains, GLM-5-Turbo offers improved execution speed and response stability.
Additional Features
In addition, the model supports a range of other technical features relevant for deployment in production systems:
- Streaming output for real-time responses
- Function calling for integration of external tools
- Context caching for efficient long-term conversations
- Structured output in formats such as JSON
- MCP integration for connecting external data sources
- Multiple thinking modes for different task types
ZClawBench: Dedicated Benchmark
To make the performance of GLM-5-Turbo measurable, Zhipu AI has developed its own benchmark: ZClawBench. This is specifically tailored to end-to-end agent tasks in the OpenClaw ecosystem and covers areas such as environment setup, software development, information retrieval, data analysis, and content creation.
“GLM-5-Turbo delivers significant improvements compared to GLM-5 in OpenClaw scenarios and outperforms several leading models in various important task categories.”
Market Positioning
With this, Zhipu AI clearly positions GLM-5-Turbo as a specialized solution for the growing market of AI agents, not as a universal competitor to models like GPT-5 or Claude. The strategy is: less breadth, more depth in one of the fastest-growing application fields in the AI industry.
Zhipu is not the only Chinese company diving into OpenClaw — there are also corresponding developments at Tencent and Alibaba.