이미지: METAL LAB 생성
Summary
- Z.ai's coding tool ZCode has reached 1 million users
- To mark the milestone, usage limits were reset for all GLM Coding Plan users
- An update was also rolled out to strengthen long-horizon task capability, aimed at seeing real engineering work through to completion
- 발표 주체
- Z.ai (즈푸)
- 발표일
- 2026-08-11
- ZCode 사용자 수
- 100만 명
- 조치 내용
- GLM Coding Plan 전체 사용자 사용량 제한 리셋
- 업데이트 방향
- 장기 작업(long-horizon) 능력을 완결된 결과물로 전환
- 관련 모델
- GLM-5.1 (2026-08-09 Hugging Face 공개, MIT 라이선스)
A coding tool now used by 1 million people
ZCode, the coding agent operated by Zhipu AI, has announced that its cumulative user count has surpassed 1 million. To mark the occasion, Z.ai said via its X account that it had reset usage limits for all users on the GLM Coding Plan. Notably, the reset applies to the entire plan's user base rather than a specific paid tier.
The GLM Coding Plan is a subscription service through which Zhipu offers its self-developed GLM-series language models for development tasks such as code generation, refactoring, and debugging. ZCode is the tool that connects this plan to actual coding work, operating by taking a user's natural-language requests and carrying out multi-step engineering tasks in sequence.
The ability to see long-horizon tasks through to completion
Alongside this announcement, Z.ai also introduced an update currently underway. The company described it as an update that "turns long-horizon task capability into completed deliverables," citing improvements in intelligence for real engineering workflows as well as cache-related enhancements. However, the exact meaning of the cache-related figures mentioned in the public post was not clearly confirmed.
Long-horizon tasks have long been considered a challenge in the coding agent industry. Unlike short question-and-answer exchanges, real development work that spans multiple files and dozens of steps often runs into problems where context is lost or earlier work is forgotten midway through. What Z.ai is emphasizing here appears to be aimed at reducing this kind of mid-task dropout and raising the proportion of tasks that are actually completed to the end.
Recent pace of GLM-series releases
Zhipu has rolled out several models in quick succession over the past few days. On August 9, it released GLM-5.1 on Hugging Face. Distributed under the MIT license, the model can be run through various frameworks including transformers, vLLM, and SGLang. This was followed on August 10 by the simultaneous release of GLM-Image, a text-to-image generation model, and GLM-OCR, an image-to-text conversion model.
| Release date | Model/Service | Description |
|---|---|---|
| Aug 9 | GLM-5.1 | Text generation, MIT license |
| Aug 10 | GLM-Image | Text-to-image generation |
| Aug 10 | GLM-OCR | Image-to-text conversion |
| Aug 11 | ZCode update | Coding agent, reached 1 million users |
Having released language, image, and OCR models in rapid succession before disclosing user metrics for its coding tool, Zhipu appears to be aiming to expand its presence across multiple fronts simultaneously rather than relying on a single model.
Zhipu releases GLM-5.1 model on Hugging Face
So what's actually changing
For existing GLM Coding Plan users, the move means they can now use the tool again without worrying about usage limits for the time being. The 1-million-user figure itself signals that Zhipu has established itself as a scale that can't be ignored in the coding agent market. As Chinese AI companies that have wielded open-source models as their weapon begin to show their presence through real service metrics rather than just competing on language model performance, how much tangible difference the next GLM-series update makes in actual development workflows will be worth watching.
