GLM-5.2
Released: June 16, 2026, according to the official announcement. (z.ai)
Positioning and capabilities
GLM-5.2 is an open-weight flagship model designed for long-horizon agent tasks, particularly software engineering. It accepts text and produces text, with a documented 1M-token context window and a maximum output of 128K tokens. It supports thinking modes, streaming, function calling and context caching. (docs.z.ai)
Its intended uses include working with large codebases, complex debugging, implementation, performance optimization and automated research. Adjustable reasoning effort lets users balance capability against latency and computational cost. The publisher reports that IndexShare reduces attention-indexing computation for long contexts. (z.ai)
The publisher makes the weights available under the MIT license and lists local inference frameworks including Transformers, vLLM and SGLang. (z.ai)
Limitations
The documented model is text-only; the 1M context and 128K output limits do not imply that every integration exposes those limits. Long-context serving requires substantial cache and infrastructure resources. Benchmark results depend on the evaluation setup and do not guarantee reliable results on every task; the publisher notes room for improvement on demanding long-horizon engineering benchmarks. (docs.z.ai)
API pricing
Z.ai international API, checked November 16, 2026: USD $1.40 per million input tokens, $0.26 per million cached input tokens, and $4.40 per million output tokens. Cached-input storage is listed as free for a limited time. These are token-based API rates, not Coding Plan subscription prices. Official pricing. (docs.z.ai)
The official mainland China pricing page was also checked, but its model-specific price could not be verified because the page required JavaScript. BigModel pricing. (bigmodel.cn)