Chinese AI startup Z.ai, known for its powerful, open source GLM family of large language models (LLMs), has introduced GLM-5-Turbo, a new, proprietary version of its open source GLM-5 model aimed at agent-driven workflows. The company bills it as a fast model for OpenGL-style tasks such as tool utilization, long-chain execution, and continuous automation.
It is now available through Z.ai’s application programming interface (API) on third-party provider OpenRouter With an approximately 202.8K-token reference window, 131.1K maximum output, and a listing price of $0.96 per million input tokens and $3.20 per million output tokens. According to our calculations, this makes it about $0.04 cheaper per total input and output cost (at 1 million tokens) than its predecessor.
| Sample |
input |
Production |
total cost |
Source |
|
grok 4.1 fast |
$0.20 |
$0.50 |
$0.70 |
xai |
|
gemini 3 flash |
$0.50 |
$3.00 |
$3.50 |
|
|
KM-K2.5 |
$0.60 |
$3.00 |
$3.60 |
moon |
|
glm-5-turbo |
$0.96 |
$3.20 |
$4.16 |
openrouter |
|
GLM-5 |
$1.00 |
$3.20 |
$4.20 |
Z.ai |
|
cloud haiku 4.5 |
$1.00 |
$5.00 |
$6.00 |
anthropic |
|
quen3-max |
$1.20 |
$6.00 |
$7.20 |
alibaba cloud |
|
gemini 3 pro |
$2.00 |
$12.00 |
$14.00 |
|
|
GPT-5.2 |
$1.75 |
$14.00 |
$15.75 |
OpenAI |
|
GPT-5.4 |
$2.50 |
$15.00 |
$17.50 |
OpenAI |
|
cloud sonnet 4.5 |
$3.00 |
$15.00 |
$18.00 |
anthropic |
|
cloud opus 4.6 |
$5.00 |
$25.00 |
$30.00 |
anthropic |
|
GPT-5.4 Pro |
$30.00 |
$180.00 |
$210.00 |
OpenAI |
Second, Z.ai is also adding the model to its GLM coding subscription product, which is its packaged coding assistant service. That service has three tiers: Lite at $27 per quarter, Pro at $81 per quarter, and Max at $216 per quarter.
Z.ai’s March 15 rollout note states that Pro customers get GLM-5-Turbo in March, while Lite customers get base GLM-5 in March and will have to wait until April for GLM-5-Turbo. The company is also taking early-access applications for enterprises through Google Forms, which suggests that some users may get access earlier than that schedule, depending on capacity.
z.ai GLM-5-Turbo is designed for “faster inference” and “deeply optimized for real-world agent workflows involving long execution chains” with improvements in complex instruction decomposition, tool utilization, scheduled and consecutive execution, and stability across extended tasks.
The release offers developers a new option for building OpenGL-style autonomous AI agents, and serves as an indication of where model vendors think enterprise demand is headed: away from chat interfaces and toward systems that can reliably execute multi-step tasks.
That’s where most of the competition is rising, as well, especially among vendors trying to win over developers and enterprise teams building internal assistants, workflow orchestrators, and coding agents.
Built for execution, not just conversation
Z.ai’s material frames GLM-5-Turbo as a model for production-like agent behavior rather than static quick-response use.
The pitch focuses on reliability in practical workflows: better command following, robust tool invocation, better handling of scheduled and persistent tasks, and faster execution in long logical chains. This positioning puts the model in the market for agents who do more than just answer questions.
The aim is to have systems that can gather information, call up tools, break down instructions and keep working through complex task sequences with little supervision.
Rather than a direct successor to GLM-5, GLM-5-Turbo appears to be a more performance-focused version: tuned for speed, device utilization, and long-chain agent stability, while the base GLM-5 remains Z.ai’s broad open-source flagship.
GLM-5-Turbo appears to be particularly competitive in OpenClaw scenarios such as information search and gathering, office and daily tasks, data analysis, development and operation, and automation. They are company supplied materials, not independent verification, but they illustrate the intended product positioning.
Background: z.ai and GLM-5 set the stage for Turbo
Founded in 2019 as a spinoff of Tsinghua University in Beijing, Z.ai – formerly Zhipu AI – is now one of China’s best-known foundation model companies. The company is headquartered in Beijing and led by CEO Zhang Peng
Z.ai listed on the Hong Kong Stock Exchange on January 8, 2026, with shares priced at HK$116.20 and opened at HK$120, with a declared market capitalization of HK$52.83 billion, making it China’s largest independent large language model developer.
As of September 30, 2025, its models were reportedly used by more than 12,000 enterprise customers, more than 80 million end-user devices, and more than 45 million developers worldwide.
Z.ai’s last major release, GLM-5, which debuted in February 2026, provides useful context for what the company is now trying to do with GLM-5-Turbo.
GLM-5 is an open-source flagship model with an MIT license, posting record-low hallucination scores on the AA-Omniscience Index, and introducing a native “Agent Mode” that can convert signals or source material into ready-to-use .docx, .pdf, and .xlsx files.
That first release was also hailed as a major technological step forward for the company. GLM-5 was scaled to 744 billion parameters with 40 billion actives per token in an expert-mixture architecture, used 28.5 trillion pretraining tokens, and relied on a new asynchronous reinforcement-learning infrastructure called “Slim” to reduce training bottlenecks and support more complex agentic behavior.
In that light, GLM-5-Turbo looks more like a replacement for GLM-5 than a narrow commercial offshoot: a version that maintains the long-context, agentic orientation of the flagship line but emphasizes speed, stability, and execution in real-world agent chains.
Developer Features and Model Packaging
On the technical side, Z.ai is packaging the GLM-5 family with the kinds of capabilities that developers now expect from serious agent-facing models, including long context handling, tools, logic support, and structured integration.
OpenRouter’s GLM-5-Turbo page lists support for tools, tool selection, and response formatting, while also revealing live performance data including average throughput and latency.
OpenRouter’s provider telemetry adds a useful deployment-level comparison between GLM-5 and GLM-5-Turbo, although the data is not completely apples-to-apples as GLM-5 is visible across multiple providers while GLM-5-Turbo is only shown through Z.ai.
On throughput, GLM-5-Turbo on OpenRouter averages 48 tokens per second, which puts it below the fastest GLM-5 endpoints shown in the screenshots, which include Fireworks at 70 toks/sec and Friendly at 58 toks/sec, but above Together’s 40 toks/sec.
On raw first-token latency, GLM-5-Turbo is slower in the available data, posting 2.92 seconds versus 0.41 seconds for Friendly’s GLM-5 endpoint, 1.00 seconds for Paracel, and 1.08 seconds for DeepInfra.
But the image end-to-end completion time improves: GLM-5-Turbo is shown at 8.16 seconds, which is faster than GLM-5 endpoint, ranging from 9.34 seconds on Fireworks to 11.23 seconds on DeepInfra.
The most notable operational benefit is in equipment reliability. GLM-5-Turbo shows a 0.67% tool call error rate, which is significantly lower than the GLM-5 providers shown, where error rates range from 2.33% to 6.41%.
For enterprise teams, this suggests a model that may not win on initial response in their current OpenRouter routing, but may still be better suited for a long-term agent where completion consistency and low tool failure matter more than the fastest first token.
Benchmarking and Pricing
A ZClawBench radar chart released by Z.ai shows GLM-5-Turbo to be particularly competitive in OpenClaw scenarios such as information discovery and aggregation, office and daily tasks, data analysis, development and operation, and automation.
Those are company-provided benchmark visuals, not independent verification, but they help explain how Z.ai wants to approach both models: GLM-5 as the broader coding and open flagship, and Turbo as the more targeted agent-execution version.
A more subtle licensing hint
One notable caveat is licensing. Z.ai says that GLM-5-Turbo is currently closed-source, but also says that the model’s capabilities and findings will be added to its next open-source model release. This is an important distinction. The company is apparently not promising to open-source the GLM-5-Turbo itself.
Instead, it is being said that the lessons, techniques, and improvements from this release will inform future open models. This makes the launch more subtle than an overtly obvious break.
Z.ai’s previous GLM strategy relied heavily on open releases and open-weight distribution, which helped it build visibility among developers.
China’s AI market may be rebalanced by moving away from open source
The licensing status of the GLM-5-Turbo also lands in the broader Chinese market context which makes the launch more notable than a simple product update.
In recent weeks, reporting around Alibaba’s Quon unit has raised new questions about how China’s leading AI labs will balance open releases with commercial pressure.
Earlier this month, Quen division chief Lin Junyang stepped down, becoming the third senior Quen executive to leave in 2026, even though Alibaba’s Quen family remains one of the most prolific open-model efforts anywhere, with more than 400 open-source models released through 2023 and more than 1 billion downloads.
Reuters then reported on March 16 that Alibaba CEO Eddie Wu would take direct control of a newly formed AI-focused business group consolidating Qian and other units, amid scrutiny over strategy, profitability and brutal price competition around open-model offerings in China.
Even without those developments being overstated, they help underscore the broader question looming over the field: whether the economics of frontier AI are beginning to push even historically open-leaning Chinese labs toward a more fragmented strategy.
This doesn’t mean that Chinese labs are abandoning open source. But the pattern is becoming harder to ignore: open models help drive adoption, developer goodwill and access to the ecosystem, while some higher-priced variants aimed at enterprise agents, coding workflows and other commercially attractive use cases may come first as proprietary products increasingly become popular.
In that sense, GLM-5-Turbo fits a larger potential shift in China’s AI market, which looks similar to the playbook used by OpenAI, Anthropic, and Google in the US: openness as delivery, proprietary systems as business.
Viewed in that light, the GLM-5-Turbo looks like much more than a speed-focused product update. This could be another sign that parts of China’s AI sector are moving toward the same hybrid model that is already common in the US: openness as distribution, proprietary systems as business.
This won’t mark the end of open-source AI from Chinese labs, but it could mean that their most strategically important agent-focused offerings appear first behind closed access, even if some of their underlying advances make their way into open releases later.
For developers evaluating agent platforms, this makes GLM-5-Turbo both a product launch and a useful signal. Z.ai is still speaking the language of open models. But with this release, it also looks like some of its most commercially relevant functions may be arriving first as proprietary infrastructure for enterprise-grade agent systems.
<a href