Z.ai Launches GLM-5.3 API: Frontier-Class Performance at Competitive Pricing

The release follows a debut last week that featured cyber capabilities, which reportedly uncovered a previously undetected vulnerability in Cursor. Developers who subscribed to a GLM Coding Plan can access the model using the OpenAI Chat Completions-compatible protocol, according to z.ai.

Pricing Structure and API Rates

Z.ai priced the GLM-5.3 API identically to its predecessor, GLM-5.2, charging $1.40 per million input tokens and $4.40 per million output tokens, according to company pricing data. Cached input costs $0.26 per million tokens, while z.ai lists cached-input storage as free for a limited time. These flat rates allow developers to transition to the new generation without facing higher per-token costs, despite z.ai claims of substantially stronger coding and long-horizon agent performance.

Using the standard comparison metric of one million input tokens plus one million output tokens, GLM-5.3 costs $5.80 in total API fees, as calculated by VentureBeat. That places the model below several high-end frontier alternatives, such as xAI’s Grok 4.6 at $8.00 for lower context rates, Moonshot AI’s Kimi K3 at $18.00, Anthropic’s Claude Opus 5 at $30.00, and OpenAI’s GPT-5.6 Sol at $35.00. However, the model is more expensive than lower-tier options, including OpenAI’s GPT-5.6 Luna at $0.20 for input and $1.20 for output, and Google’s introductory Gemini 3.7 Flash rate of $0.75 input and $3.75 output through December 31, 2026.

Independent Benchmarks and Verbosity Analysis

Independent evaluations place GLM-5.3 among the top open-weights models available. Artificial Analysis assigned GLM-5.3 a score of 60 on its Intelligence Index, tying it with Moonshot AI’s Kimi K3 as the highest-performing open-weights model globally and marking a seven-point improvement over GLM-5.2.

At the same time, Artificial Analysis estimated the cost of GLM-5.3 at approximately $0.68 per Intelligence Index task, compared to roughly $0.44 for GLM-5.2. That cost discrepancy arises because GLM-5.3 is more verbose than its predecessor, meaning flat per-token API prices do not translate to identical workload costs for completed tasks, according to Artificial Analysis.

Open Weights Roadmap

Z.ai stated plans to make the model weights for GLM-5.3 openly available for download, though the company has not yet announced a precise release date or licensing terms. Until open weights are published, developer access remains restricted to the API and existing subscription plans.