ZAug 19, 2026 23:23 UTCZ.ai Begins Offering GLM-5.3 via API
Chinese startup Z.ai has begun offering API access to its large language model 'GLM-5.3'. Pricing remains unchanged from the previous generation GLM-5.2, with input tokens at $1.40 per million and output tokens at $4.40 per million. While Z.ai has indicated plans to open-source the model weights in the future, specific timing and licensing details remain undecided.
Chinese startup Z.ai has begun offering its open-source large language model 'GLM-5.3' to developers via API. Developers can now integrate GLM-5.3 into their own applications and AI agents through this API. GLM-5.3 is the new model following the previous generation GLM-5.2, and Z.ai emphasizes significant improvements particularly in coding performance and agentic behavior for long-duration tasks.
GLM-5.3 debuted just last week. At its launch, it was reported to have discovered previously undetected vulnerabilities in Cursor (a coding assistance tool) in the context of security research, with its cybersecurity performance garnering some attention. This API launch marks the transition to a stage where developers can actually use the model in a practical form. Note that developers who previously contracted GLM Coding Plan are currently limited to support for the OpenAI Chat Completions-compatible protocol only.
Pricing remains unchanged from GLM-5.2. Input tokens are offered at $1.40 per million, and output tokens at $4.40 per million. For cached input, the rate is $0.26 per million tokens, and cached input storage is offered at no cost for a limited period. This means developers can migrate to new model generations without incurring additional costs.
Compared to APIs of other major models, GLM-5.3's pricing positions it in the upper tier. In a straightforward comparison of input and output at 100 million tokens combined, Meta's Muse Spark 1.1/1.2 totals $5.50, xAI's Grok 4.6 (under 200k prompt tokens) is $8.00, Anthropic's Claude Opus 5 is $30.00, and OpenAI's GPT-5.6 Sol (Fast mode) is $70.00, with GLM-5.3's combined $5.80 positioned between these. Meanwhile, some models from DeepSeek and Xiaomi are offered at lower price points than GLM-5.3, indicating intense price competition.
Z.ai has indicated its intention to open-source the model weights, but specific timing and licensing details remain unclear. Open-weight models differ significantly from commercial API usage in that developers can run models on their own infrastructure. How these details unfold in future announcements will be a key point of attention for many developers.
In the current generative AI model market, companies are competing fiercely on both performance and pricing. The entry of Chinese startups like Z.ai into the global developer market through API services can broaden options for AI services while potentially exerting downward pressure on overall pricing. How well GLM-5.3 accumulates track records in coding and agentic applications will become a key axis for future evaluation.
This article is an original work independently written and edited by the AI issue editorial team based on factual reporting. © AI issue. Unauthorized reproduction, redistribution, or use for AI training is prohibited.