DeepSeek Price Surge Meets Resistance: Intellect Spectrum GLM-5.3 Enhanced, Peak Input Price More Competitive

According to DynaCheck Beating monitoring, DeepSeek has just announced a significant price increase for V4, while SmartSpec GLM-5.3 has promptly followed suit.

Assuming GLM-5.3 maintains the same API price as GLM-5.2, the peak prices of the common tokens on both sides have basically equalized. For GLM-5.3, the price is $8 for input and $28 for output per million tokens; V4-Pro is at $9 and $27. GLM's input is $1 cheaper, and DeepSeek's output is $1 cheaper.

After price equalization, the performance gap becomes even more critical. According to the calibration table published by SmartSpec, there are 9 items that can be directly compared between the two models, with GLM-5.3 winning in 7 items. DeepSWE is at 66.9 versus 62.7, Agents' Last Exam is at 28.5 versus 25.7, HLE with Tools is at 62.5 versus 60.0, and CyberGym is at 84.5 versus 83.3. DeepSeek only leads in NL2Repo and Toolathlon Verified. Terminal Bench 2.1 has basically equalized on both sides, at 88.2 and 87.9 respectively.

Therefore, after this price increase, DeepSeek's advantage in running complex Coding Agents during peak hours has become awkward: V4-Pro is no longer much cheaper, and GLM-5.3 is actually stronger in most comparable evaluations. Especially in scenarios requiring continuous code writing, tool invocation, and execution of long tasks, GLM-5.3 is now more attractive.

The truly cheaper aspects of DeepSeek are still in off-peak hours and caching. V4-Pro's off-peak input and output prices have decreased to $4.5 and $13.5, directly 50% off; cache hits are even lower at $0.15, reaching only $0.3 even at peak times, while GLM is at $2. Running high-difficulty Agents during the day is more cost-effective with GLM-5.3; running off-peak, or tasks that heavily reuse cache, DeepSeek remains much cheaper.