Beijing‑based Z.ai released GLM-5.2 in May, and the model quickly climbed to fourth place on Artificial Analysis's Intelligence Index v4.1 with a score of 51. It outranked all open‑weight competitors, beating MiniMax‑M3, DeepSeek V4 Pro and Kimi K2.6, and landed second on Code Arena's front‑end coding leaderboard, ahead of Anthropic's Claude Opus variants.

Pricing sets GLM-5.2 apart. Z.ai charges $1.40 per million input tokens and $4.40 per million output tokens on its own API. Third‑party hosts list even lower rates. Compared with Claude Opus or OpenAI's GPT‑5.5, the cost per output token is roughly one‑fifth to one‑seventh, depending on the provider and tier. The cheap price tag has rattled rivals that rely on higher‑priced, closed‑source offerings.

David Sacks, who served as the White House AI czar under former President Donald Trump, praised the model as "as good as the currently available models from OpenAI and Anthropic," placing it just below Anthropic's Opus 4.8 and roughly level with GPT‑5.5. Beyond headline scores, GLM-5.2 topped Artificial Analysis's GDPval‑AA v2 metric with a 1,524 rating, edging out MiniMax‑M3 and DeepSeek V4 Pro and nearing GPT‑5.5's top reasoning setting.

Analysts note a trade‑off: GLM-5.2 generates more output tokens per task than other open‑weight models, which can erode its cost advantage in real‑world use. Still, the model's capabilities and low price have attracted attention from developers seeking affordable, high‑performing AI.

Another distinguishing factor is the hardware behind GLM-5.2. Z.ai trained and runs the model on an estimated 100,000 Huawei Ascend 910B processors, a wholly domestic Chinese silicon stack that avoids Nvidia, AMD, or Intel components. The approach directly counters U.S. export controls that aim to limit Chinese access to advanced chips.

The model's weights are released under an unrestricted MIT licence, allowing any organization to download, modify and run the system locally for the cost of electricity alone. This openness provides a stark contrast to the proprietary models dominating the U.S. market.

The timing aligns with a shifting regulatory environment. Washington recently ordered Anthropic to suspend its Fable 5 and Mythos 5 models for foreign users, and the Trump administration asked OpenAI to stagger the rollout of its next generation model. Those moves have left a gap that GLM-5.2 is poised to fill, especially for users barred from the latest U.S. offerings.

Z.ai's stock has reflected the excitement. The Hong Kong‑listed company debuted in January, and its shares surged after GLM-5.2's launch, climbing several multiples above the IPO price. Investors appear to be betting on the firm's ability to capitalize on both performance and price advantages.

Z.ai has hinted at a successor, GLM-5.5, expected in August, though no firm date has been confirmed. If the upcoming model maintains the current trajectory, Chinese AI firms could further narrow the gap with their U.S. counterparts while operating under domestic hardware constraints.

Questo articolo è stato scritto con l'assistenza dell'IA.
News Factory APP - notizie agentiche per potenziare il tuo SEO e AEO.