The AI ​​price war continues: model services move from "luxury goods" to "bulk utilities"

The price war for AI model services continues to heat up, and the industry is moving from a "luxury product" that a few people can afford to a "bulk utility" that everyone can afford - resources like water, electricity, and cloud resources are consumed on a large scale and at low cost. The two latest actions come from the two camps of OpenAI and DeepSeek respectively.

Even the top closed sources have joined in the price reduction

OpenAI While launching the GPT-5.6 series, it has reduced the prices of some products and actively participated in price competition. This in itself is a signal: even top closed-source manufacturers have to respond to the price pressure brought by open source and domestic models. The price war is no longer divided into camps, and the competition extends from "model capabilities" to "cost, stability and supply chain."

DeepSeek’s unified API mapping

DeepSeek Unifiedly maps the API to the thinking/non-thinking mode of deepseek-v4-flash to simplify the calling form. This step is a combination of "simplification + price reduction": on the one hand, it reduces developer access costs, on the other hand, it dilutes inference costs through a single high-performance model, achieving the coexistence of service standardization and low price. For domestic small and medium-sized developers, low price + unified API + open source ecosystem are making domestic models the default choice, further promoting the innovation and prosperity of the application layer.

"Public utility" is a sign of industry maturity

When model services are consumed on a large scale and at low cost like water and electricity, it means that the AI infrastructure process is being completed. Subsequently, the business model has evolved from "charge by call" to "infrastructure utility", and the focus of profits has shifted to the application layer and computing power layer. For small and medium-sized model manufacturers, the crowding out effect in this trend also needs to be vigilant. The competitive threshold of the industry is being raised by the price war itself.

Several directions worth tracking in the future:

1. Price reduction transmission: The specific extent of the GPT-5.6 series price reduction and the transmission to the pricing of competing products.
2. Actual performance: The actual price and performance of deepseek-v4-flash after unifying the API.
3. crowding out effect: the crowding out effect of price war on small and medium-sized model manufacturers.
4. Long-term trend: The long-term trend of reasoning costs and computing power supply in the context of "public utility".