**[DeepSeek Price Hike Hits a Snag: Zhipu GLM-5.3 is Stronger with Better Peak Pricing]** According to monitoring by 动察 Beating, just as DeepSeek announced a significant price hike for V4, Zhipu GLM-5.3 has conveniently caught up. Assuming GLM-5.3 retains the API pricing of GLM-5.2, the regular token prices during peak periods are now almost identical. GLM-5.3 charges ¥8 per million tokens for input and ¥28 for output, while V4-Pro charges ¥9 and ¥27, respectively. GLM is ¥1 cheaper for input, while DeepSeek is ¥1 cheaper for output. With prices now leveled, the performance gap becomes more critical. Based on the like-for-like comparison table released by Zhipu, the two models can be directly compared across nine metrics, with GLM-5.3 winning in seven of them. For example, DeepSWE scores are 66.9 vs. 62.7, Agents' Last Exam scores are 28.5 vs. 25.7, HLE with Tools scores are 62.5 vs. 60.0, and CyberGym scores are 84.5 vs. 83.3. DeepSeek only leads in NL2Repo and Toolathlon Verified. For Terminal Bench 2.1, the two are nearly tied at 88.2 and 87.9, respectively. As a result, after DeepSeek's price hike, its advantage in running complex Coding Agents during peak periods becomes awkward: V4-Pro is no longer significantly cheaper, and GLM-5.3 outperforms it in most like-for-like evaluations. Particularly in scenarios requiring continuous coding, tool invocation, and executing long tasks, GLM-5.3 is now more appealing. DeepSeek's real cost advantage still lies in off-peak periods and caching. During off-peak hours, V4-Pro's input and output prices drop to ¥4.5 and ¥13.5, effectively halving the cost. Cache hits are even cheaper at ¥0.15, and only ¥0.3 during peak periods, compared to GLM's ¥2. For running high-difficulty Agents during the day, GLM-5.3 is more cost-effective; however, for off-peak operations or tasks that heavily reuse cached data, DeepSeek remains significantly cheaper. [Original Link]