Chinese AI models close gap with Anthropic and OpenAI as GLM-5.2 tops open-weight rankings

Z.ai's Latest Model Shakes Up AI Rankings

Beijing-based startup Z.ai has released GLM-5.2, an open-weight AI model that has rapidly climbed to fourth place on Artificial Analysis Intelligence Index v4.1 with a score of 51, positioning it ahead of established competitors like MiniMax-M3, DeepSeek V4 Pro, and Google's Gemini 3.1 Pro Preview5

5

. Released on June 13th under a permissive MIT license, the model has sparked what some experts are calling a "mini DeepSeek moment," drawing comparisons to the Chinese startup that shocked markets in early 2025 with its cost-effective approach to AI development2

2

. Within a week of launch, GLM-5.2 had topped openly available leaderboards, and Z.ai's market value surpassed HK$1 trillion, approximately US$128 billion1

1

.

The timing proved significant. Just one day before Z.ai's release, the U.S. Commerce Department issued an export-control directive barring Anthropic from supplying Fable 5 or Mythos 5 to any foreign national, forcing the company to disable both models worldwide1

1

. This Fable 5 ban created a gap in the market that GLM-5.2 quickly filled, becoming the most capable model many users outside the U.S. could legally access1

1

. David Sacks, former AI czar under President Donald Trump, noted that "we now have a Chinese open-weight model that is as good as the currently available models from OpenAI and Anthropic," describing it as sitting just below Anthropic's Opus 4.8 and roughly level with GPT-5.52

2

5

5

.

Cost-Effective AI Model Disrupts Pricing Structure

The AI model performance benchmarks tell only part of the story. GLM-5.2's pricing structure represents a fundamental challenge to Western frontier labs. Z.ai charges $1.40 per million input tokens and $4.40 per million output tokens on its first-party API, with third-party hosts listing it even lower5

5

. Multiple analyses peg this at somewhere between a fifth and a seventh of what Claude Opus or GPT-5.5 cost per output token, though exact ratios vary by provider and tier5

5

. Jefferies strategist Christopher Wood wrote in a report to clients that GLM-5.2 "is almost equal to Anthropic as a competitor for the corporate market and is just one quarter of the cost in terms of cost per token"3

3

.

This cost advantage arrives as corporate America shifts from an era of "tokenmaxxing"—allowing developers to spend on AI without restraint—to a focus on efficiency and return on investment3

3

. Flo Crivello, CEO of AI startup Lindy, switched his company entirely off Anthropic's Claude models to DeepSeek, saying "we did it, and you could see that cost curve go down, like, crash to the ground"3

3

. The model's MIT license allows anyone to download, modify, and self-host its weights, eliminating API costs entirely for companies with sufficient infrastructure1

1

5

5

.

Trained Entirely on Huawei Ascend 910B Chips

GLM-5.2's training stack represents a direct challenge to US export controls on AI. Z.ai has been on the U.S. Entity List since January 2025, cutting it off from Nvidia's H100, H200, and B200 accelerators1

1

. The company states that the GLM-5 family was trained on roughly 100,000 Huawei Ascend 910B processors using the MindSpore framework, with no Nvidia silicon at any stage1

1

5

5

. The export controls on advanced AI chips were designed to prevent precisely this kind of result, yet they have evidently failed to do so1

1

.

However, hardware parity remains elusive. The Ascend 910C sits at roughly 60% of an Nvidia H100's inference performance, according to a December report from the Council on Foreign Relations, with a wide gap on efficiency and cluster scale1

1

. The same report projects that by as early as next year, the best U.S. chips could be more than 17 times more powerful than Huawei's top parts1

1

. While GLM-5.2 demonstrates that a frontier-class open model can be produced on a fully domestic stack, model parity does not equal hardware parity1

1

.

AI Industry Competition Intensifies Amid Geopolitical Tensions

On Code Arena's front-end coding rankings, GLM-5.2 holds second place behind Fable 5, demonstrating strong agentic capabilities—the ability to execute complex tasks with minimal prompting2

2

5

5

. The model scored 62.1 on SWE-bench Pro, compared to GPT-5.5's 58.6, and took first place on Design Arena's human-preference coding board, finishing roughly 10 Elo points ahead of Fable1

1

. On Artificial Analysis's GDPval-AA v2 metric, which measures real-world agentic work rather than abstract reasoning, it scores 1,524, effectively tied with GPT-5.5 on its highest reasoning setting5

5

.

Yet the model trails on certain benchmarks. On Artificial Analysis's AA-Briefcase test, which scores multi-week knowledge tasks built from thousands of fragmented inputs, Fable 5 led with 1,587 Elo, followed by Opus 4.8 at 1,356, and GLM-5.2 in third place at 1,266, before the export ban took Fable out of contention1

1

. It also lags on raw terminal work, scoring 81.0 on Terminal-Bench 2.1 against Opus 4.8's 85.0 and GPT-5.5's 84.01

1

.

Data Security Concerns and Enterprise Adoption

One major hurdle to widespread adoption remains data security concerns that have limited use of Chinese AI models by U.S. enterprises, particularly in regulated industries like banking and cybersecurity2

2

. Corporate filings show that several of Z.ai's shareholders are controlled by a Chinese government agency that supervises the country's defense industry4

4

. Wei Sun, principal AI analyst at Counterpoint Research, noted that "in the EU and U.S., some clients, partners and regulated industries may simply be unwilling to accept Chinese models in their AI stack, regardless of technical performance or price"2

2

.

Despite these concerns, the technological gap is narrowing. A report by non-profit RAND found that Chinese LLMs' global market share jumped to 13% from 3% in the two months after DeepSeek launched its R1 model in January 20252

2

. Six of the models now on leading AI leaderboards were developed in China4

4

. Venture capitalist Marc Andreessen wrote that "many smart people/AI insiders are saying GLM-5.2 is the first Chinese AI model to match and often beat the American big lab public AI models with no compromises"3

3

.

The geopolitical implications extend beyond benchmarks. "The international developer community is increasingly aware that relying solely on proprietary, U.S.-based API models carries significant risk," said Brian Tse, founder and CEO of Concordia AI, a Beijing-based consultancy focused on AI safety2

2

. As U.S. authorities grapple with how to respond, Chinese AI models continue gaining ground through open-weight distribution, lower token consumption costs, and improved performance across coding and agentic tasks. Z.ai founder Tang Jie, responding to Elon Musk on X, suggested the company could produce a model on par with Anthropic's Fable before the first quarter of next year, adding "won't take that long"3

3

.