Zhipu, GLM-5.2 and DeepSeek-V4 Pro continue to jockey, domestic LLM API prices keep rising
Analysis on August 11 shows Zhipu GLM-5.2 pricing up ~83% in 2026 cumulatively; July coding plan entry tier rose from 39 to 118 RMB; DeepSeek V4 Pro delayed.
On August 11 analysis noted that Zhipu has completed three API price increases in 2026 so far, with overall pricing up around 83% versus end of 2025; at the end of July its coding subscription plan entry tier rose from 39 RMB to 118 RMB. This contradicts rumors of a Zhipu GLM 5.2 95% price cut — no official announcement supports that claim in reality.
Background: on July 31 DeepSeek released the V4-Flash GA version with significantly strengthened agent capabilities; the average cost per benchmark test is only 3 cents, far below Anthropic Claude Fable 5 at $3.15 and OpenAI GPT-5.6 Sol at $1.86. OpenRouter data shows the DeepSeek-V4-Flash GA logged 8.83 trillion tokens that week, a 570% weekly jump, topping the global ranking on extreme low-price strategy.
Against rising demand, DeepSeek preannounced an across-the-board API price increase on August 6, expected to be substantial, with the market generally expecting a September rollout and increases of 35%-50%. DeepSeek V4 Pro is also delayed, with analysis attributing this to compute supply-demand tensions — current infrastructure cannot simultaneously support low-margin Flash public testing and high-margin Pro deployment.
Alibaba is also pushing commercialization segmentation. A Reuters report on August 7 said Alibaba plans to introduce revenue sharing for large commercial users of Qwen3.8-Max — weights remain open, but large commercial users must share a portion of revenue with Alibaba. Kaiyuan Securities interprets the DeepSeek price increase as reflecting demand growth, inference capacity, and cost recovery pressure; Alibaba's revenue share extends commercialization from API fees to commercial deployment in customer-owned data centers.
A three-line commercialization pattern is taking shape for domestic models: low-price acquisition, API monetization, and revenue-share lock-in of large customers. Zhipu has built a 1GW domestic AI compute cluster with over 50,000 domestic compute chips currently running; the long-standing GLM Coding Plan purchase restrictions have ended and new pricing has been raised significantly.