Tongyi Qianwen significantly reduced its price, with the cost of the AI model API dropping by 97%.

On May 21, the API call prices for the Tongyi Qianwen commercialization model and Open Source model have been significantly reduced for developers. Among them, the input price for the GPT-4 level market maker model Qwen-Long has dropped from 0.02 yuan/k tokens to 0.5 yuan/million tokens, a decrease of 97%.

On May 9th, Tongyi released the Open Source model Qwen1.5-110B with 110 billion parameters. This model surpassed similar 70 billion parameter models in multiple benchmark evaluations and ranked first in the leaderboard of open source large models.

The combination strategy of “price reduction + Open Source” is becoming a consensus among global large model manufacturers. This helps to address the two major pain points faced by AI application developers: the high cost of model APIs and the insufficient quality of open source models, thereby promoting the widespread implementation of AI applications.

Recently, several large model manufacturers have successively launched low-priced products or price reduction measures. For example, a quantitative company’s open source MoE model API is priced at only about one percent of a well-known model. Another AI company has reduced the calling price of its personal version model from 5 yuan to 1 yuan per million tokens. Additionally, some companies have launched new versions of models with comparable performance but at half the price. Furthermore, some companies have even announced that their market maker models will be completely free or permanently open for API access. This series of price reductions may stem from advancements in large model inference technology and a decrease in costs, objectively providing developers with more choices and facilitating the development of AI applications.

In addition to price reductions, Tongyi has also launched eight large language models with parameters ranging from 500 million to 110 billion, as well as multi-modal models for vision, audio, code, and more. Smaller models are suitable for deployment on mobile phones, PCs, and other endpoints, while larger models can support enterprise-level and research-level applications, and medium-sized models strike a balance between performance and efficiency. This multi-size, multi-modal product matrix helps to meet the application needs of different scenarios.

The general reduction in the prices of large model APIs reflects the intensifying competition in the industry and the cost reductions brought about by technological advancements. This will provide more options for AI application developers and is expected to promote the rapid development and implementation of AI applications. At the same time, the application of large models on the client side is also worth noting. In the future, with the continuous evolution of large model technology and the advancement of the commercialization process, AI applications are expected to flourish in a wider range of fields.