Zhipu has raised prices for the third time this year, increasing by another 10%, as domestic large models collectively abandon price wars.

Ask AI · What are the driving factors behind the collective abandonment of price wars by domestic large models?

Zhipu raises prices again. On April 8, Zhipu officially released and open-sourced its GLM-5.1 model. The model aggregation platform OpenRouter shows that Zhipu GLM has increased its price by another 10%. After the price adjustment, the cache hit token price for GLM-5.1 in coding scenarios has approached the level of Claude Sonnet 4.6 under Anthropic. As of press time, Zhipu (2513.HK) surged over 14% intraday, trading at HKD 888.5, with a market capitalization of HKD 396.1 billion.

Zhipu recently raised prices three times, with another 10% increase

This is already Zhipu’s third price hike recently. After launching its new flagship model GLM-5 on February 12, Zhipu previously adjusted its GLM Coding Plan package, with an overall increase of at least 30%. On March 16, Zhipu announced the world’s first general large model optimized for the “Lobster” scene, GLM-5-Turbo, and raised the API prices for GLM-5-Turbo by 20%.

On April 8, the model aggregation platform OpenRouter showed that Zhipu GLM4 increased its price by another 10%. Currently, the input price for GLM-5 Turbo is $1.20 per million tokens, and the output price is $4 per million. The latest released GLM-5.1 has an input price of $1.26 per million and an output price of $3.96 per million. Calculated this way, GLM-5.1’s input price has increased by 5% compared to the previous generation, while the output price has decreased slightly by 1%.

The latest GLM-5.1 continues to lead in programming capabilities. In the combined average scores of the SWE-bench Pro, Terminal-Bench, and NL2Repo code evaluation benchmarks, it ranks third globally, first domestically, and first in open source; in the SWE-bench Pro benchmark closest to real software development, it even achieved the first domestic model to surpass Opus 4.6, setting a new global best.

In addition, unlike current models mainly interacting in minutes, GLM-5.1 can work continuously and autonomously for up to 8 hours in a single task, which is the biggest highlight of this model release. During task execution, the model can autonomously plan, execute, and test, actively switch strategies when hitting obstacles, self-repair after errors, and ultimately deliver complete engineering-level results.

Zhipu recently released its first annual report since going public. The financial report shows that by March this year, Zhipu’s API call prices increased by 83% compared to the end of last year, yet demand remains high. Zhipu CEO Zhang Peng stated that there are many companies providing API services based on computing power in the market, and customers’ acceptance of price increases and continued choice indicates that the model’s capabilities are indeed leading. “I believe that the long-term pricing is fundamentally determined by value. Resources that effectively replace human labor, improve conversion efficiency, and enhance intelligence are scarce and valuable. We focus more on the value created per token and the value brought to customers. Only when customers recognize this value are they willing to pay higher costs to use it. I believe that pricing power is still determined by our technological strength and the leading position brought by long-term trends.”

Demand surges, domestic large models collectively abandon price wars

Unlike the narrative of price wars among domestic large model vendors a year ago, this year, domestic model companies have collectively started raising prices. In March alone, Tencent Cloud, Alibaba Cloud, and Baidu Intelligent Cloud each announced price adjustments, raising prices for AI computing services and large model-related products. Among them, Tencent Cloud’s Mixyuan series models increased by up to 463.13%, Alibaba Cloud’s Pingtouge Zhenwu 810E and other computing cards increased by 5%–34%, and Baidu Cloud’s AI computing services increased by 5%–30%. Several previously free beta models have also ended their promotional periods and shifted to commercial billing.

This is partly due to the explosive demand driven by the popularity of various intelligent agents. An insider revealed that Alibaba Cloud’s AI computing power, storage, and other products saw price increases of up to 34%, mainly because of a surge in token calls. “During the Spring Festival, AI agent applications exploded. Alibaba Cloud MaaS business Bailian achieved its highest growth rate from January to March this year.”

MaaS API is a common business model chosen by model vendors. This model provides cloud-based AI large model call services through standardized interfaces. In a previous earnings call, Zhipu CEO Zhang Peng mentioned that AI capabilities have shifted from being usable and fun to solving increasingly complex and important problems, turning token API calls and consumption into real economic value. The emergence of new applications like “Lobster” and the anticipation of device-level native intelligence also suggest that future API and token consumption will grow exponentially.

He also said that after several years of development, the large model industry has been seeking a simple, cost-effective business model to accelerate growth. API is a way to convert AI infrastructure capabilities into resources for economic operation.

During Tencent’s Q4 and full-year 2025 earnings call, management responded to the price increases in the domestic cloud computing market, noting that CPU and other infrastructure capacities have long been fully booked. Suppliers tend to prioritize the largest and most stable customers, while smaller companies and cloud service providers no longer believe they can reliably secure supply chain support and have to turn to ultra-large resource providers. These ultra-large providers operate at low profit margins, and when demand rebounds, the entire industry has little choice but to raise prices.

Reporting: Nandu N Video Reporter Lin Wenqi