The GLM-5.3 model from Chinese company z.ai is now available through an application programming interface (API), allowing developers to integrate it into applications and software agents instead of using it directly only through the company’s plans. The launch comes at a price similar to that of GLM-5.2, despite z.ai announcing major improvements in coding and long-horizon task execution capabilities.
Pricing and access
z.ai charges $1.40 per million input tokens and $4.40 per million output tokens. Cached input tokens cost $0.26 per million tokens, while the company is offering free storage for cached inputs for a limited period.
Keeping the price unchanged compared with GLM-5.2 means that the difference between the two generations does not appear in the advertised price per token. However, this does not necessarily mean that the total cost of a task will be identical, because the bill is affected by the volume of inputs and outputs, the use of caching, and the number of tokens the model produces.
Users who previously subscribed to the GLM Coding Plan currently have access limited to the protocol compatible with OpenAI Chat Completions. z.ai also said it intends to make the model weights openly available, but the company has not specified an exact date and has not disclosed the license that will govern this use.
How does the price compare with other models?
According to VentureBeat’s simplified comparison, which assumes the use of one million input tokens and one million output tokens, GLM-5.3 costs $5.80. The same comparison puts Grok 4.6 at $8 at the lower context rate, Kimi K3 at $18, Claude Opus 5 at $30, and GPT-5.6 Sol at $35.
These figures illustrate the price range more than they provide an estimate of an actual operating bill. GLM-5.3 is not the cheapest among capable models; Google offers a promotional price for Gemini 3.7 Flash of $0.75 for input and $3.75 for output through December 31, 2026, while OpenAI’s GPT-5.6 Luna costs about $0.20 for input and $1.20 for output. Nevertheless, z.ai’s model remains cheaper than a number of advanced model APIs with which it is compared.
What does this mean for developers?
The immediate practical benefit is that a development team can test a new model for coding or for running software agents without increasing the advertised token price compared with GLM-5.2. This could make it easier to compare performance and cost within applications that rely on repeated API calls.
However, the fixed price alone is not enough to assess efficiency. Artificial Analysis gave GLM-5.3 a score of 60 on the Intelligence Index, the same score achieved by Kimi K3, making them the two highest-scoring open-weight models according to the cited comparison. GLM-5.3 also exceeded GLM-5.2’s score by seven points.
By contrast, Artificial Analysis estimated the cost per task within the index at about $0.68 for GLM-5.3, compared with approximately $0.44 for GLM-5.2, despite the identical token price on the API. This is related to GLM-5.3 being more verbose than its predecessor, which may lead it to consume more tokens to complete the same task.
The model’s first appearance was reportedly associated with advanced cyber capabilities that were said to have helped it find a previously undiscovered vulnerability in Cursor. The text provides no additional details to verify this result, so this point remains a reported claim rather than a sufficient basis for an independent security evaluation of the model.