While Washington cuts off foreign access to Anthropic’s top models, China delivers its answer as a free download: Beijing-based AI developer Z.ai – formerly known as Zhipu AI – has released its new flagship, GLM-5.2, putting forward what independent measurements rank as the most capable openly available language model currently on the market. After an initial rollout to paying coding customers on June 13, the company released the model weights under the permissive MIT license – no usage restrictions, no regional locks.
The top spot is confirmed by independent analysis house Artificial Analysis. On the Intelligence Index v4.1, GLM-5.2 scores 51, placing it ahead of MiniMax-M3 (44), DeepSeek V4 Pro (44) and Kimi K2.6 (43). On this more agentically focused index, GLM-5.2 even pulls ahead of Google’s top models: Gemini 3.1 Pro Preview lands at 46 here, and the fast Gemini 3.5 Flash at 50. One detail with a caveat, though – these are different model classes, and earlier versions of the index rated the Gemini models higher in places.
Even more telling are the human-rated leaderboards from Arena.ai (LMArena), which are based on anonymous head-to-head comparisons and are considered hard to game. On the Code Arena, GLM-5.2 ranks second with 1,595 points – making it the strongest model currently usable at all, since Anthropic’s Claude Fable 5 was removed from the ranking following the US export ban. On the frontend list, GLM-5.2 even places ahead of Claude Opus 4.7 and Opus 4.8 in thinking mode.
On top of that come first place on the Design Arena and, by a wide margin, the top spot among open models on the Agent Arena. On Artificial Analysis’ real-world metric GDPval-AA v2, the model reaches roughly 1,524 points – putting it on par with OpenAI’s GPT-5.5.
Strong, but not flawless
GLM-5.2 doesn’t quite reach the closed frontier in coding. On SWE-bench Pro, the model scores 62.1 points, seven points behind Claude Opus 4.8 (69.2). On several long-horizon coding benchmarks, however, the gap to Opus 4.8 shrinks to around a single percentage point – a remarkable result for a freely downloadable model.
The real lever: price
Technically, GLM-5.2 is a mixture-of-experts model with 744 billion total and 40 billion active parameters – the same size as its predecessor, but with a context window quadrupled to one million tokens. The decisive difference, however, lies in the price: via providers such as OpenRouter, the model costs around $1.40 per million input and $4.40 per million output tokens, while GPT-5.5 sits at $5/$30 and Claude Opus at $5/$25. The open alternative thus undercuts the US flagships many times over – the central selling point in the price-sensitive “value-for-money” segment, where Chinese providers are increasingly gaining ground.
The launch doesn’t fall into a tense week by accident. The Trump administration ordered Anthropic’s most advanced models, Fable 5 and Mythos 5, to be blocked for foreign nationals – and on the same day, Zhipu announced the open-source release of GLM-5.2 with no usage restrictions. For developers outside the US, GLM-5.2 is therefore the most capable openly licensed model currently available. The catch: anyone using Z.ai’s cloud API is subject to Chinese law – a point that falls away with pure self-hosting of the MIT weights.
On the markets, the combination of model launch and geopolitical tailwind went down well. The listed Zhipu entity, Knowledge Atlas Technology (HKEX: 2513), gained sharply: the stock jumped as much as 48 percent at the start of the week after JPMorgan raised its price target from 950 to 1,400 Hong Kong dollars and named the stock an AI winner; Bank of America initiated coverage with a buy recommendation and a price target of 1,250 HK dollars. The share most recently traded at around 1,559 HK dollars – equivalent to a market capitalization of roughly 650 billion HK dollars.
That keeps the stock at a multiple of its issue price. Zhipu went public on the Hong Kong exchange on January 8, 2026, at 116.20 HK dollars and marked its all-time high of 1,993 HK dollars on May 29 – followed by a sharp correction and now the rebound around GLM-5.2. The risk that remains is the expiry of the first lock-up period: from July 8, the shares of the first cornerstone investors become tradable, abruptly multiplying the freely available float. In parallel, Zhipu is preparing a secondary listing on Shanghai’s STAR Market with a targeted volume of around 15 billion yuan.
GLM-5.2 marks the point at which an open model becomes a serious argument against the closed frontier – not because it wins everywhere, but because it comes close enough while being significantly cheaper and freely available. At the very moment US providers are restricting their access, a measure of bargaining power shifts toward open source. And one thing is also clear: the “China threat” isn’t just called DeepSeek anymore – it’s also called Z.ai.
Aus Datenschutz-Gründen ist dieser Inhalt ausgeblendet. Die Einbettung von externen Inhalten kann in den Datenschutz-Einstellungen aktiviert werden: