Z.ai

Zhipu AI was founded as a spin-out from Tsinghua University by researchers Tang Jie and Li Juanzi, developing GLM (General Language Model) technology using an autoregressive blank infilling training strategy.

Zhipu AI released GLM-130B, a 130-billion parameter bilingual (English/Chinese) language model trained on a novel General Language Model architecture. The model matched GPT-3 performance on several benchmarks while being fully open-source, establishing Zhipu as a leading Chinese AI lab in the global open-weight model ecosystem.

Zhipu AI released ChatGLM-6B, an open-source bilingual dialogue model optimized for Chinese-English conversations that could run on consumer GPUs. ChatGLM rapidly became one of the most popular open-source LLMs in China, powering thousands of downstream applications and driving the country's open-source AI ecosystem.

Zhipu AI raised $300 million in Series B funding at a $1.5 billion valuation, backed by investors including Alibaba and Tencent. The funding round reflected growing investor confidence in China's domestic AI model ecosystem amid tightening US chip export restrictions and positioned Zhipu as a national AI champion.

Zhipu AI secured approximately $400 million in funding with a valuation of about $3 billion, with participation from Saudi Arabia's Prosperity7 Ventures, marking significant international investment in a Chinese AI startup.

Zhipu released GLM-4-Plus, its most capable model to date, competing with top-tier models from OpenAI and Anthropic in the Chinese market.

Zhipu AI released two GLM models under the free and open-source MIT License, signaling a strategic shift toward open-source AI development and broader accessibility of its foundation models.

Zhipu released GLM-4.5 and GLM-4.5 Air next-generation language models and rebranded itself internationally as Z.ai, positioning for global expansion.

Z.ai held its IPO on the Hong Kong Stock Exchange, becoming China's first major LLM company to go public. The IPO was considered a milestone for China's AI industry.

Z.ai (formerly Zhipu AI), the Tsinghua University-born AI lab behind the GLM model family, went public on the Hong Kong Stock Exchange (HKEX: 02513.HK), becoming the first publicly listed Chinese AI company. The IPO marked a milestone for China's AI industry, providing the company with public market capital to fund its aggressive GLM model development cadence, which included the GLM-5 744B-parameter MoE model released one month later in February 2026.

Z.ai released GLM-5, its next-generation flagship model, continuing to compete with top-tier international models while maintaining competitive API pricing.

Z.ai released GLM-5.1 as open-source after initially offering it to subscription users in late March. The company increased API prices by 10% while maintaining lower prices than Anthropic's Opus 4.6, and its share price increased 11.5% on release.

Z.ai launched GLM-5.2 featuring a one-million token context window, a fivefold increase from GLM-5.1's 200K-token window. This positioned Z.ai alongside frontier models in long-context capabilities.

Z.ai (Zhipu AI) launched ZCode, a developer harness built on its GLM-5.2 model, designed to power autonomous coding assistants rivaling Anthropic's Claude Code and Cursor. The move expands Z.ai's product line beyond model releases into the agentic coding tools market.

Z.ai (Zhipu AI) released GLM-5.3, an open-weight 743B model using the same base as GLM-5.2 with all gains from post-training. Achieved 50 percent improvement on Z.ai Code Bench and a 6x jump on Terminal-Bench 3 (from 4.6 to 28.3). Notably, post-training produced exploit-chain reasoning capabilities the company did not intend, finding 1,097 critical vulnerabilities in Linux, WebKit, and FreeBSD during testing. Weights released under an open license.

Twelve days after GLM-5.3, Z.ai released GLM-5.3-Flash, a 320B-parameter open-weight mixture-of-experts model with 18B active parameters and a one-million-token context window -- the first natively multimodal model in the GLM-5 series, priced at roughly 0.15 USD per million output tokens. Before launch, Z.ai tested the model anonymously as 'ox-alpha' on OpenCode and OpenRouter, where it became the most popular model of the week; all of that preview traffic was served on domestic Chinese AI accelerators, a milestone in China's push for inference independent of US export-controlled hardware. The release reinforced Z.ai's position as the leading Chinese open-weights lab, following its first-ever two-week safety hold on the GLM-5.3 weights earlier in the month.

Z.ai published GLM-5.3 open weights on Hugging Face, ending an unprecedented two-week hold for what the company called its most extensive risk review to date, driven by cyber capability that emerged faster than expected. The delay broke Z.ai's own precedent of shipping MIT-licensed weights within days of launch (GLM-5.2). In the same window, Z.ai confirmed that Ox Alpha, a stealth model running anonymously on OpenRouter, was in fact GLM-5.3-Flash -- 320B total parameters (18B active), MIT-licensed, and served entirely on Chinese-made AI chips. The move marked a broader industry shift in which 'launched' and 'downloadable' became separate events for cyber-capable frontier models.

Z.ai (formerly Zhipu AI) completed a Hong Kong financing of about 5 billion USD net (up to HK$39.27 billion), pairing a roughly 2 billion USD share placement of up to 22 million new H shares with about 3 billion USD of zero-coupon convertible bonds. Filings showed most net proceeds directed to next-generation AI model development, self-training research, and compute infrastructure -- the company's second major raise in two months after its July capital raise, and one of the largest AI financings by a Chinese lab. The raise deepens the capital gap between Z.ai and US frontier labs amid the intensifying US-China AI race, though shares fell over 10 percent on the dilution.

Developers discovered Z.ai's coding assistant ZCode had been silently packaging entire project workspaces, including complete Git histories and cached credentials, and uploading them to Alibaba cloud storage without consent; one researcher counted 564 upload attempts of a 313 MB encrypted archive whose decryption key was held only on Z.ai's servers. Z.ai apologized, promised to open-source ZCode for third-party security audits, and offered compensation, but the episode triggered a formal legal demand in China and a lasting trust crisis for the GLM model maker.