[News] Rising Costs and Demand Drive China’s LLM Price Jump: Zhipu GLM‑5 Hikes 30% in First 2026 Increase

Last Lunar New Year, DeepSeek made headlines worldwide with its V3 model, putting Chinese AI firmly on the global map. This year, as the holiday season approaches, Chinese tech giants are back in the spotlight, igniting a fresh surge in the AI race.

Among the moves drawing attention, Zhipu AI announced two major developments. According to chinastarmarket.cn, the firm’s next-gen flagship large model, GLM‑5, debuted in overseas markets, while it also issued a GLM Coding Plan price adjustment notice, which marks the first significant price hike for domestic large language models in 2026.

The report highlights that Zhipu AI’s package prices are rising by at least 30%, while existing subscribers will continue paying their original rates. The changes took effect immediately on February 12, it adds.

Notably, chinastarmarket.cn suggests that overseas GLM‑5 pricing increases are even higher than domestic rates: Coding Plan subscription prices rise 30%–60%, and API usage fees jump 67%–100%.

Riding this trend, the report notes that China’s AI industry has long relied on low prices and subsidies to grab market share. Since 2025, however, more companies have reportedly halted price cuts, leading to stabilized API pricing across the board, with some products seeing significant increases.

The chinastarmarket.cn report also names the price hike moves from the so-called “Six Rising Tigers” of large models in China, adding that Zhipu, Moonshot AI, MiniMax, and StepFun had already raised certain API prices in 2025. While Baichuan Intelligence and 01.AI kept prices unchanged, major tech firms such as Alibaba, ByteDance, Tencent, and Baidu have broadly adopted tiered pricing strategies, the report notes.

Why China’s LLMs Are Getting Pricier

The report highlights the drivers behind the price hike, using Zhipu’s GLM‑5 as an example. On the demand side, soaring usage is the main factor. The GLM Coding Plan, focused on developer code generation and programming assistance, has seen rapid growth in both users and API calls, keeping the platform under constant high load.

On the cost side, rising investment in computing power and technology is driving price adjustments, the report notes, adding that Zhipu explicitly noted it is “simultaneously increasing computing resources and optimizing the model.”

More Chinese AI Models Poised to Hit the Market

The market is bracing for another wave from China’s tech giants, with new models arriving over the holiday season. According to Anue, DeepSeek will unveil its next-generation flagship V4 in mid-February. A recent paper by founder Liang Wenfeng and industry sources notes that V4 introduces two core architectural innovations—mHC and Engram—designed to cut training and inference costs while challenging current leaders in programming capabilities, the report adds.

Anue explains that Engram allows the model to store static knowledge—like entities and fixed expressions—in low-cost DRAM instead of expensive high-bandwidth GPU memory (HBM). During inference, the model quickly retrieves this data, freeing GPU power for complex calculations and reducing China’s dependence on HBM.

On the other hand, Commercial Times notes that ByteDance, which earlier drew global attention with its Seedance 2.0 audio-visual generation model, would further update its flagship Doubao model on February 14, acknowledging that Seedance 2.0 is “still far from perfect.” The report notes that the release will include Doubao 2.0, Seedance 2.0, and the image-generation model Seedream 5.0 Preview as well.

Read more

• [News] DeepSeek Expands Context Tenfold as Zhipu Rolls Out New Model in China’s AI Race
• [News] China Steps Up AI Push: Alibaba Latest Qwen Model Nears U.S. Rivals; Moonshot Advances

(Photo credit: FREEPIK)