Since DeepSeek’s debut early last year, which jolted markets by pairing low cost with notable capability, the AI landscape has often been framed as a choice between inexpensive Chinese models with more limited features and heavily funded U.S. offerings from OpenAI and Anthropic. That binary is being tested again by GLM-5.2, a model launched last month by Beijing startup Z.ai, also known as Zhipu AI.
GLM-5.2 has generated a flurry of attention for its capacity to handle coding tasks and to run agent-style workflows - executing multi-step assignments with minimal prompting - in ways that some observers judge to be nearly on par with U.S. leaders while costing far less. The model has climbed third-party usage charts, including ranking above Anthropic’s models on platforms such as OpenRouter, and has attracted public praise from figures ranging from Snowflake CEO Sridhar Ramaswamy to venture capitalist Marc Andreessen.
On the All-In podcast, David Sacks, former U.S. administration AI official, described GLM-5.2 as evidence that "we now have a Chinese open-weight model that is as good as the currently available models from OpenAI and Anthropic." He added that "it is just a tick below Opus 4.8 (from Anthropic) and right up there with GPT 5.5 (from OpenAI)," and warned "we cannot afford to do things that slow our companies down." These remarks came in the days before Washington moved to lift curbs on Anthropic’s Fable and Mythos models on Tuesday.
Observers link GLM-5.2’s sudden popularity in part to recent developments affecting U.S. models, including the curbs on Anthropic and a delayed public rollout of OpenAI’s GPT-5.6. For some industry participants, that uncertainty has increased demand for viable non-U.S. alternatives.
"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 AI safety consultancy. That sentiment dovetails with heightened interest in lower-cost open-source approaches, particularly as organizations confront rising and at times unpredictable bills for closed-source agentic AIs, which typically consume large volumes of tokens, the units used to measure AI usage.
GLM-5.2’s market metrics underscore the attention it has attracted. The model is listed fifth on Artificial Analysis’ large language model intelligence leaderboard, which measures performance across benchmarks including reasoning and coding. It places second on Code Arena’s front-end coding rankings, a measure of how effectively models generate websites and front-end applications. Independent observers note that GLM-5.2 runs at roughly one-sixth the cost of closed U.S. frontier offerings such as Anthropic’s Claude and the GPT series.
Z.ai has not disclosed its development spending for GLM-5.2. The company declined to comment when approached. Anthropic and OpenAI did not immediately reply to requests for comment.
In a public exchange on X last month with Elon Musk, Z.ai founder Tang Jie said the startup could produce a model comparable to Anthropic’s Fable before the first quarter of next year. Commentators who work with open-source tools note a practical change in GLM-5.2’s release: it functions as a more plug-and-play product than many predecessors.
"The shift GLM-5.2 brings is that the open-source model has become a plug-and-play, out-of-the-box product," said Tiezhen Wang, formerly the APAC lead at Hugging Face. "You just deploy the model and without doing any complex fine-tuning systems, it is in a highly usable, ready-to-use state. This drastically lowers the barrier to entry for open-source adoption." That accessibility helps explain why some segments of the market, particularly developers and smaller firms, gravitate toward these options.
Barriers to broader enterprise uptake
Despite the technical and cost advantages cited by proponents, GLM-5.2 faces well-documented hurdles before it can be widely embedded in enterprise AI stacks, especially among U.S. firms in regulated sectors. Data security concerns are a primary constraint. Many large corporations and organizations in highly regulated fields such as banking and cybersecurity remain cautious about integrating models developed in China.
Migration and upgrading of enterprise AI systems is not instantaneous. "The migration and upgrading of enterprise AI systems typically takes several months," said Tiezhen Wang, highlighting the time it takes for thorough testing, integration, and risk assessment.
Wei Sun, principal AI analyst at Counterpoint Research, noted that in the EU and U.S. some customers, partners and regulated industries may be unwilling to accept Chinese models in their AI stack, "regardless of technical performance or price." Those reservations point to a segmentation in adoption patterns: while major corporations move cautiously, startups and small- and medium-sized enterprises are progressing more quickly in experimenting with and adopting alternative models.
Usage data from an earlier market study echoes that divergence. A RAND report released earlier this year, which analysed website traffic across 135 countries, found that Chinese large language models’ global market share rose to 13% from 3% in the two months after DeepSeek released its R1 model in January last year. That growth was most pronounced among developing countries and states with closer political and economic links to Beijing.
Some experts argue that safety concerns about Chinese models can be mitigated operationally - for instance, by running the models on U.S. cloud providers or on a firm's own on-premises servers - but that reassurance has not erased the caution of large regulated enterprises.
For developers and many smaller organisations, practical considerations like performance, cost and deployment reliability matter more than a model’s origin. "Developers tend to care less about where a model comes from than whether it works, how much it costs and whether they can deploy or access it reliably," said Poe Zhao, China tech analyst and founder of the Hello China Tech newsletter. He expects a pattern of partial routing - where some workloads shift to alternatives - rather than an overnight replacement of OpenAI or Anthropic models. In this view, GLM-5.2 represents a "mini DeepSeek moment" concentrated in developer ecosystems rather than a wholesale industry upheaval.
As GLM-5.2 gains traction among developers and gets added to third-party platforms, the episode highlights a set of competing forces shaping AI adoption: cost and usability pull companies toward cheaper, ready-to-deploy open-source models, while security concerns, regulatory environments and the inertia of enterprise migration continue to limit how quickly that shift can proceed at scale. How companies balance those pressures will determine whether GLM-5.2’s early momentum translates into sustained market share beyond developer communities and smaller firms.