How does GLM-5.3 compare to competitors in terms of cost and performance?

GLM-5.3 stands out with its low token pricing ($1.4/$4.4 per million) and strong performance, rivaling top models like Kimi K3. It also supports data caching to reduce costs.

When will z.ai release the GLM-5.3 model weights?

The company has announced plans to open-source the model weights but has not yet disclosed the exact timeline or licensing terms.

← All news

Artificial intelligence

Chinese z.ai Launches GLM-5.3 API at $1.4 per Million Tokens

August 20, 2026

Photo: VentureBeat

Quick answer

Chinese startup z.ai launched the GLM-5.3 language model API with token pricing at $1.4/$4.4 per million, undercutting many competitors while maintaining high performance.

Chinese startup z.ai has launched an API for its next-generation GLM-5.3 language model. Previously recognized for uncovering an unknown vulnerability in the Cursor tool, the model is now available to developers for integration into applications and autonomous agents. Currently, the API supports only the OpenAI Chat Completions-compatible protocol, though the company plans to open-source the model weights in the future.

GLM-5.3’s pricing remains consistent with its predecessor: $1.4 per million input tokens and $4.4 per million output tokens. Cached input data costs just $0.26 per million tokens, while storage is temporarily free. This makes GLM-5.3 one of the most affordable high-performance models, despite claimed improvements in programming and long-term task planning.

In comparison to competitors, GLM-5.3 sits in a mid-range pricing tier. For example, Google’s Gemini 3.7 Flash* offers rates of $0.75/$3.75 (until the end of 2026), while OpenAI’s GPT-5.6 Luna charges $0.20/$1.20. However, GLM-5.3 delivers strong value for money: according to Artificial Analysis, it scored 60 points in the Intelligence Index, matching Kimi K3 and surpassing GLM-5.2 by 7 points.

Experts note that while GLM-5.3’s token pricing is competitive, real-world usage may cost more due to its higher verbosity compared to its predecessor. Nonetheless, it provides developers with another affordable option for testing cutting-edge capabilities in programming and autonomous agents.

Common questions

How does GLM-5.3 compare to competitors in terms of cost and performance?: GLM-5.3 stands out with its low token pricing ($1.4/$4.4 per million) and strong performance, rivaling top models like Kimi K3. It also supports data caching to reduce costs.
When will z.ai release the GLM-5.3 model weights?: The company has announced plans to open-source the model weights but has not yet disclosed the exact timeline or licensing terms.
How did GLM-5.3 perform in benchmarks?: According to Artificial Analysis, GLM-5.3 scored 60 points in the Intelligence Index, matching Kimi K3 and outperforming its predecessor, GLM-5.2, by 7 points.

Dzen feed: /feed/dzen.xml · RSS: /feed.xml