Zhipu Releases GLM-5.2, a 744-Billion-Parameter Open Coding Model, Days After Washington Cut Off Foreign Access to Anthropic's Claude

Overview

Beijing-based Zhipu AI released GLM-5.2, a 744-billion-parameter open-weight model aimed at long-horizon coding, days after the US government cut off foreign access to Anthropic’s flagship models. The Chinese lab announced the full release on Saturday and said the weights would be formally open-sourced under the permissive MIT license, according to the South China Morning Post. The launch landed in a market reshaped by a US export-control order issued days earlier against Anthropic.

What We Know

GLM-5.2 is the latest model in the GLM-5 series. Per the project’s official repository, the model “scales from 355B parameters (32B active) to 744B parameters (40B active)” relative to its GLM-4.5 predecessor, a Mixture-of-Experts design in which only a fraction of the network is used per token, as documented in the GLM-5 repository. The same documentation states the model carries “a solid 1M-token context that stably sustains long-horizon work” and that pre-training data was increased “from 23T to 28.5T tokens.”

On architecture, the repository says GLM-5 “integrates DeepSeek Sparse Attention (DSA), largely reducing deployment cost while preserving long-context capacity,” and adds a technique the team calls IndexShare, which “reuses the same indexer across every four sparse attention layers, reducing per-token FLOPs by 2.9× at a 1M context length,” according to the GLM-5 repository. The documentation reports coding benchmark results of 62.1 on SWE-bench Pro and 81.0 on Terminal-Bench 2.1, and lists deployment support across SGLang, vLLM, Transformers and KTransformers, per the same repository.

The South China Morning Post reports that GLM-5.2 will be available to all users of Zhipu’s new GLM Coding Plan subscription, “which is priced at just a tenth of Anthropic’s premium Claude Code and Claude Max tiers,” and that the model’s API “was scheduled to go live this week,” the outlet reports.

The Export-Control Backdrop

The release followed an abrupt US move against Anthropic. On June 12, 2026, Commerce Secretary Howard Lutnick sent a letter to Anthropic chief executive Dario Amodei ordering the company to restrict its newest models, according to MarkTechPost. The directive barred access to Claude Fable 5 and Mythos 5 by any foreign national inside or outside the United States, a scope that extended even to Anthropic’s own foreign-national employees, the same report notes. Unable to filter foreign users from US users in real time, Anthropic disabled both models, MarkTechPost reports. Anthropic characterized the underlying issue as a small number of minor, already-known vulnerabilities comparable to those found in other public models, according to the same report.

Zhipu’s announcement “came shortly after US AI giant Anthropic abruptly suspended access to its flagship models, Fable-5 and Mythos-5,” the South China Morning Post reports. Investors read the sequence as an opening for Chinese open-weight challengers. Shares of Zhipu AI “soared on Monday after it released GLM-5.2,” surging “as much as 48 per cent to HK$1,620 in morning trading” and ending the day “up 32.8 per cent at HK$1,457,” according to the South China Morning Post.

What We Don’t Know

Zhipu has not published a full third-party benchmark suite alongside the launch beyond the figures listed in its repository, leaving independent verification of GLM-5.2’s coding claims to outside evaluators. The exact date the open weights and API would become broadly available remained described only as “this week” at the time of the stock move, per the South China Morning Post. It also remains unclear whether the US restrictions on Anthropic’s models are temporary or will harden into a longer-term licensing regime, and how that uncertainty will shape enterprise adoption of open-weight alternatives.

Analysis

The timing places GLM-5.2 squarely in the gap created by Washington’s order. Where access to a leading closed model was curtailed for non-US users, a frontier-class open-weight model arrived under a permissive license that lets organizations download and run it on their own hardware. The market reaction in Hong Kong suggests investors see export controls aimed at constraining one ecosystem as a potential tailwind for the open-weight competitors positioned to absorb displaced demand.