News · Z.ai (Zhipu) · 2 min read
Event date: · Published:
Z.ai releases open-weight GLM-5.3 and reports a 50% gain on Z.ai Code Bench over GLM-5.2
On August 18, 2026, Z.ai released GLM-5.3, a text-only reasoning model with 1M context, and reported a 50% gain on Z.ai Code Bench over GLM-5.2. It was available on all GLM Coding Plans, and its weights were published on Hugging Face.
Archive entry: this covers an event from August 18, 2026, and we published it on November 8, 2026. The details come from sources dated at the time and may have changed since.
On August 18, 2026, Z.ai released GLM-5.3, a text-only reasoning model built on the same base model as GLM-5.2. The release appeared in the Z.ai release notes, and Z.ai reported a 50% gain over GLM-5.2 on Z.ai Code Bench.
What Z.ai reported
- Z.ai Code Bench: a 50% gain over GLM-5.2. At max effort, GLM-5.3 scored 34.5% with about 75K output tokens, versus 23.4% with 96K for GLM-5.2.
- Other benchmarks: compared with GLM-5.2, Terminal-Bench 3.0 rose from 4.6 to 28.3, DeepSWE v1.1 from 46.2 to 66.9 and Agents' Last Exam from 23.8 to 28.5. Z.ai called GLM-5.3 state of the art among open-source models on Terminal-Bench 3.0 and Agents' Last Exam (CLI).
- Code review: Z.ai said it matches Mythos 5 in white-box code review, finding 2,436 vulnerabilities, 1,097 of them medium or high severity.
- Model details: text-only, with 1M context and 128K maximum output. Reasoning was always on, with low, high or max effort (default max), and requests with thinking disabled failed.
Access and availability
- Endpoints: OpenAI Chat Completions at
https://api.z.ai/api/coding/paas/v4, OpenAI Responses athttps://api.z.ai/api/v1and Anthropic athttps://api.z.ai/api/anthropic. - Plans: GLM-5.3 was available on all GLM Coding Plans. The points-based plan used 50% of the points for off-peak calls, including all day at weekends.
- Weights: on Hugging Face as
zai-org/GLM-5.3under the glm-5.3 license.
What it means for people building with AI
GLM Coding Plan subscribers could use a model that Z.ai said improved Z.ai Code Bench results by 50% over GLM-5.2. Through the Anthropic endpoint, clients in the style of Claude Code could use it, and the weights were published on Hugging Face.
Other platforms and sources
Mistral's platform made GLM-5.3 generally available on September 28, 2026, according to the Mistral changelog. More details are in the GLM-5.3 guide and on the GLM-5.3 Hugging Face page.
Sources
- New Released (Z.ai (Zhipu), August 18, 2026, official)
- GLM-5.3 (Z.ai (Zhipu), official)
- zai-org/GLM-5.3 (Z.ai (Zhipu), official)
- Changelog | Mistral Docs (Mistral AI, September 28, 2026)
Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.