GLM-5-Turbo Released by Zhipu: A Practical Foundation Model Optimized for Lobster Intelligence

Zhipu AI has officially released the new foundation model GLM-5-Turbo. This model has been deeply restructured for the "Lobster" (OpenClaw) scenario of open-source agents, aiming to solve the problem of "stalling" in long-chain and complex tasks for general large models, evolving AI from a simple efficiency tool into a round-the-clock digital workforce.

Underlying Evolution: Breaking Long Task Stalls

To make the model more "capable of work," the development team reconstructed the training data through real agent workflows, focusing on enhancing four core capabilities:

Stable Tool Calls: Enhance the accuracy of external Skills calls, ensuring no breakdowns at critical stages.
Complex Instruction Decomposition: Accurately identify goals and plan steps, supporting efficient multi-agent collaboration.
Time Dimension Awareness: Optimize for scheduled triggers and long-running scenarios, ensuring task continuity.
High Throughput Execution: Improve the execution efficiency of long-term business flows, leading in response stability among similar models.

Real-World Testing: Outstanding Performance

Zhipu also released the agent evaluation benchmark ZClawBench. In a blind test covering diverse fields such as code development, data analysis, and content creation, the new model with the code name Pony-Alpha-2 received the preference of 90% respondents. Feedback from multiple testing teams including Alibaba, Meituan, Kuaishou, and ByteDance showed that it demonstrates first-tier competitive advantages in tool call accuracy, instruction following, and "hallucination" control in office scenarios.

Terminal Adaptation and Security Management

In terms of ecosystem deployment, GLM-5-Turbo has been adapted to the "Lobster Box" by SoftBank's MechRevolution, offering a native AI hardware experience. At the same time, Zhipu launched the enterprise-level security system "Claw for Enterprise Security," which enables complex multi-agent collaboration to move away from the "black box" era through visual monitoring, permission orchestration, and audit logs, ensuring data stays within the enterprise boundary.

To alleviate the pressure of computing costs, Zhipu introduced the "Lobster Package" including personal and team versions. Currently, GLM-5-Turbo has been launched on the Zhipu Open Platform, and users can enjoy limited-time benefits: during off-peak hours, the quota consumption coefficient is only 1x, and the promotion will continue until the end of April.

Coding Plan Max: https://z.ai/subscribe

OpenRouter: https://openrouter.ai/z-ai/glm-5-turbo

API: https://docs.z.ai/guides/llm/glm-5-turbo

Safety Warning After the Popularity of Lobster: Ministry of State Security Releases OpenClaw Safety Breeding Manual

In 2026, the open-source AI tool OpenClaw (nickname "Lobster") became a phenomenon due to its strong autonomous execution capability. As the trend of breeding "Lobsters" spread, its inherent security risks caught the attention of the state. The Ministry of State Security released the "Lobster" Safety Breeding Manual, reminding users to distinguish and use it in a standardized way. The core advantage of OpenClaw lies in its ability to generate solutions and execute them autonomously, which differs from traditional large models that only provide suggestions.

NVIDIA Launches NemoClaw to Promote the Enterprise-Level Intelligent Agent Strategy

NVIDIA CEO Huang Renxun announced the enterprise-level AI intelligent agent platform NemoClaw at GTC 2026. The platform is built on the open-source framework OpenClaw, with the core purpose of providing "enterprise-level armor" for OpenClaw, focusing on solving security and privacy issues when enterprises deploy AI intelligent agents locally. Huang Renxun emphasized that enterprises should elevate the OpenClaw strategy to the same important infrastructure strategic level as Linux and Kubernetes.

Bridging the Dimension: Xiao Du Skills Launches on ClawHub, Entering the Lobster Era of Smart Home

The integration of Xiao Du Smart Home Skills with ClawHub marks the official inclusion of its ecosystem into the OpenClaw open ecosystem. Previously, Xiao Du had completed the MCP upgrade of its smart home control capabilities, enabling intelligent entities to interact with the physical world, transforming from a voice assistant into a programmable intelligent entity platform.