TECH

Z.AI Pushes AI Model With Coding Edge to Rival Anthropic, Open AI

Published: Aug. 14, 2026 6:24 p.m. GMT+8

00:00

00:00/00:00

Listen to this article 1x

Z.AI's GLM-5.3 uses the same base model as its predecessor, GLM-5.2, but the firm substantially scaled post-training. Photo: VCG

Chinese AI developer Z.AI Co. Ltd. released a new version of its flagship model Friday, betting that intensive post-training can boost coding and cybersecurity performance without building a larger underlying model.

The Hong Kong-listed company said GLM-5.3 uses the same base model as its predecessor, GLM-5.2, but the firm substantially scaled post-training. Z.AI said the approach raised performance on long-running coding and agent tasks.

You've accessed an article available only to subscribers

Subscribe today for just $.99.

VIEW OPTIONS

Unlock exclusive discounts with a Caixin group subscription — ideal for teams and organizations.

Subscribe to both Caixin Global and Bloomberg - for the price of one.

Share this article

Open WeChat and scan the QR code

DIGEST HUB

Digest Hub

Back

Explore the story in 30 seconds

Z.AI released GLM-5.3 using the same base model but scaled post-training, boosting coding and cybersecurity performance.
It showed 50% improvement over GLM-5.2 in internal tests and topped open-source peers in benchmarks like Terminal Bench 3.0.
In cybersecurity, it matched some top models but underperformed in deep reasoning. Annual recurring revenue hit $1 billion in July.

AI generated, for reference only

Who’s Who

Z.AI Co. Ltd.: Z.AI Co. Ltd., a Hong Kong-listed Chinese AI developer, released GLM-5.3, using intensive post-training on its existing base model to boost coding and cybersecurity performance. It claims to be the most capable open-source programming model. The company plans to release the model's weights soon and reached $1 billion in annual recurring revenue in early July.

Anthropic: The article mentions Anthropic's models Fable 5 and Claude Mythos 5. Z.AI claimed its GLM-5.3 matched Mythos 5 in cybersecurity code review and approached/exceeded Fable 5 on benchmarks, but underperformed top North American closed-source models in deep reasoning.

OpenAI: The article mentions OpenAI's GPT-5.6 Sol model. Z.AI claimed its GLM-5.3 approached or exceeded it in programming benchmarks and matched it in some cybersecurity tasks. The article also notes GLM-5.2 was used by Hugging Face to patch a vulnerability caused by an OpenAI model attack.

Hugging Face: Hugging Face, an AI community platform, used Z.AI's GLM-5.2 model on July 21 to patch a vulnerability caused by an OpenAI model attack. It selected this model because other advanced closed-source models had safety restrictions that blocked the patching request.

AI generated, for reference only

What Happened When

Before 2026：: Z.AI Co. Ltd. existed as a Hong Kong-listed company.

Early July 2026：: Z.AI's annual recurring revenue reached $1 billion, growing approximately 15 times from the beginning of 2026, hitting the annual commercialization target ahead of schedule.

July 21, 2026：: Hugging Face used the previous GLM-5.2 model to patch a vulnerability caused by an OpenAI model attack.

August 14, 2026 (Friday)：: Z.AI Co. Ltd. released GLM-5.3, a new version of its flagship model, with scaled post-training. The company claimed it outperformed the predecessor GLM-5.2 and topped open-source peers in benchmarks.

AI generated, for reference only