According to a research report released by Huatai Securities, the price reduction of GPT-5.6 has shifted the competitive landscape among large AI models. The focus has moved from solely evaluating capability rankings to considering 'equivalent intelligence costs' as a crucial factor. On July 30th, OpenAI announced a 20% price cut for Terra and an 80% reduction for Luna. The following day, the newly launched DeepSeek V4 Flash 0731 achieved a score of 50 on the Artificial Analysis Intelligence Index, just one point shy of Luna's score. In terms of cost, DeepSeek V4 Flash offered a hybrid pricing model, with an average task cost of approximately $0.06 per million Tokens and $0.03 for specific tasks. These costs represent reductions of about 65% and 57%, respectively, compared to Luna's pricing.

Huatai Securities noted that Kimi K3 has showcased the impressive capability potential of domestic open-weight models, earning a score of 57. Meanwhile, DS V4 Flash has set a new cost benchmark within the 50-point capability range. Chinese AI models have adopted a competitive strategy that combines 'strong capabilities' with 'high cost-effectiveness.' This approach is expected to fuel the growth of downstream AI applications. Therefore, Huatai Securities recommends that investors pay close attention to the investment themes surrounding AI applications and domestic AI models.