z.ai launches GLM-5.3 API, offering frontier coding model at unchanged token rates
Chinese startup z.ai has released its advanced GLM-5.3 language model via API at the same per-token rate as its predecessor, giving developers a lower-cost alternative to premium models for complex agent workloads.
Chinese artificial intelligence startup z.ai has made its GLM-5.3 language model available via application programming interface. The release follows the model’s debut last week, which featured advanced cybersecurity capabilities that reportedly identified a previously undetected vulnerability in the coding tool Cursor.
The company is maintaining the same pricing structure as its predecessor, GLM-5.2. Input tokens are priced at $1.40 per million, while output tokens cost $4.40 per million. Cached input is available for $0.26 per million tokens, with cached-input storage currently listed as free for a limited time.
This pricing positions GLM-5.3 well below several high-end frontier models. A standard comparison of one million input and one million output tokens totals $5.80 for GLM-5.3. By contrast, the same workload costs $8 for Grok 4.6, $18 for Kimi K3, $30 for Claude Opus 5, and $35 for GPT-5.6 Sol.
However, GLM-5.3 is not the lowest-cost capable model on the market. Google currently offers introductory pricing for Gemini 3.7 Flash at $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026. OpenAI’s GPT-5.6 Luna is priced even lower at $0.20 for input and $1.20 for output.
Despite the flat pricing, the actual cost per completed task has risen. Artificial Analysis estimates GLM-5.3 costs roughly $0.68 per Intelligence Index task, up from $0.44 for GLM-5.2. The research firm attributes this increase to GLM-5.3 being more verbose than its predecessor, demonstrating that flat per-token rates do not guarantee flat workload costs.
The model’s performance credentials remain strong in independent testing. Artificial Analysis assigns GLM-5.3 a score of 60 on its Intelligence Index. This ties it with Kimi K3 as the top-performing open weights model globally and marks a seven-point improvement over GLM-5.2.
For developers, the API launch provides a relatively low-cost avenue for testing frontier-class coding and long-horizon agent workloads. z.ai has stated it plans to make the model’s weights openly available, though a precise release date and licensing terms have yet to be announced. Currently, developers on a GLM Coding Plan remain limited to the OpenAI Chat Completions-compatible protocol.
