Beijing-Based Z.ai Launches GLM 5.2 Open-Weight Model, Eyes 'Mythos-Class' AI Capabilities by Early 2027

Beijing-based Z.ai, formerly known as Zhipu AI, is making significant strides in the global artificial intelligence arena, with its leadership signaling China's aggressive pursuit of 'Mythos-class' AI models within the next six to nine months. This push for advanced AI capabilities is underscored by the company's recent release of its flagship open-weight model, GLM 5.2, which has garnered attention for its robust performance.

Zhang Peng, CEO of Z.ai, has consistently highlighted the rapid progress of China's AI ecosystem, aligning with industry predictions that the nation could achieve 'Fable-level' (a term popularized by Elon Musk to describe highly advanced AI) or 'Mythos-class' AI sophistication by the first quarter of 2027. Indeed, Elon Musk recently stated, "Maybe Q1," when asked about China reaching 'Fable-level' AI. An X user, Teortaxes, further speculated that a complete "PRC Mythos ('Fable')" – referring to a Chinese equivalent of Anthropic's high-tier models – could potentially launch between November and December 2026. DeepSeek V4 Pro Price Cut: Chinese AI Startup Permanently Lowers Flagship Model API Rates by 75% to Challenge Global Competitors.

GLM 5.2: A New Frontier in Open-Weight AI

The launch of GLM 5.2 by Z.ai marks a pivotal moment in the competitive landscape of large language models. Officially released on June 16, 2026, for public access, following an initial rollout to GLM Coding Plan users on June 13, 2026, GLM 5.2 is designed for complex, long-horizon tasks, particularly in software engineering. The model is available as open-weight under an MIT license, making its weights publicly accessible for developers and enterprises globally.

Key Specifications of GLM 5.2:

• Developer: Z.ai (formerly Zhipu AI), Beijing, China
• Release Date: June 16, 2026 (official public release)
• Model Type: Open-weight, MIT License
• Parameters: 753 billion (Mixture-of-Experts with ~40 billion active parameters)
• Context Window: 1 million tokens (usable)
• Output Capacity: Up to 131,072 tokens
• Training Hardware: Based on the GLM-5 lineage, which leveraged domestic Huawei Ascend chips.
• Core Capabilities: Autonomous software engineering, long-horizon coding tasks, multi-token prediction (MTP), strong in design and creative visual understanding.

Challenging the Global AI Leaders

GLM 5.2 is already making waves by demonstrating impressive performance benchmarks. It has reportedly surpassed OpenAI's GPT 5.5 on certain design benchmarks. Furthermore, it ranks as the strongest open-source model on coding benchmarks, scoring 81.0 on Terminal-Bench 2.1 and 62.1 on SWE-bench Pro, significantly improving on its predecessor, GLM 5.1. The model also narrows the performance gap with closed-source frontier models like Anthropic's Claude Opus 4.8, trailing it by a mere 1% on the FrontierSWE long-horizon coding benchmark.

Zixuan Li, Director of Product and genAI Strategy at Z.ai, has previously highlighted the company's focus on coding and agentic capabilities, aligning with GLM 5.2's strengths. The competitive pricing of GLM 5.2, starting at $1.40 per million input tokens and $4.40 per million output tokens via FriendliAI, also positions it as a cost-effective alternative to proprietary Western models.

The 'Mythos-Class' Vision for China

The term 'Mythos-class' AI generally refers to highly advanced, autonomous AI systems capable of complex reasoning and operational autonomy, exemplified by models like Anthropic's Claude Fable 5 and Claude Mythos 5, which are being rolled out with enhanced safety mechanisms for high-risk domains. Telegram Ban in India: Users Flock to VPNs and Alternative Apps as Government Blocks Platform Over Exam Fraud Concerns, Says Report.

Z.ai's trajectory with its GLM series suggests a clear ambition to develop models that can compete at this elite level, further cementing China's position as a formidable player in the global AI arms race. The rapid iteration of models from Z.ai and other Chinese tech giants indicates a determined effort to not only match but potentially surpass existing benchmarks in critical AI applications.

Disclaimer: AI tools assisted in compiling the foundational data and research for this report. The final content was reviewed, edited and verified by human editors at LatestLY.

(The above story first appeared on LatestLY on Jun 19, 2026 10:01 AM IST. For more news and updates on politics, world, sports, entertainment and lifestyle, log on to our website latestly.com).