The price war of large models escalates, Tongyi Qianwen lowers prices again

Recently, Alibaba Cloud announced a significant reduction in the API call prices for the Tongyi Qianwen commercial and open-source models. Among them, the input price for the main model Qwen-Long, at the GPT-4 level, decreased from 0.02 yuan/k tokens to 0.5 yuan/million tokens, a drop of up to 97%.

Previously, Alibaba Tongyi released the open-source model Qwen1.5-110B with a parameter scale of 110 billion. This model has performed excellently in multiple benchmark evaluations, surpassing similar models with a scale of 7 billion parameters. On the HuggingFace leaderboard for open-source large models, Qwen1.5-110B has taken the top spot.

The combination strategy of "price reduction + open source" is becoming a consensus among global large model manufacturers. This approach helps address two major pain points faced by AI application developers: the high cost of large model APIs and the inadequate quality of open source models. Lowering the barriers to use will promote the widespread implementation of AI applications.

In fact, Tongyi Qianwen is not the first large model to adopt a price reduction strategy. Since May, several large model products have successively introduced low-price plans or announced price cuts:

1. The DeepSeek under Huanfang Quantitative has launched the open-source second-generation MoE model, with API pricing only one percent of a well-known large model.

2. Zhipu AI has reduced the calling price of the personal version of the GLM-3Turbo model product from 5 yuan / million tokens to 1 yuan / million tokens.

3. A well-known AI company has launched a new version of its model, which performs at par with the flagship version but is priced at half and has doubled the call speed.

4. The pricing for the Beanbag Master Model inference input is only 0.0008 yuan/k Tokens.

5. Baidu Intelligent Cloud announced that its Wenxin large model's two main versions are completely free.

6. iFLYTEK announces that the iFLYTEK Spark Lite API will be permanently open for free.

The price war for large model APIs is intensifying, likely stemming from advancements in inference technology and decreasing costs. This not only offers developers more choices but will also drive the vigorous growth of AI applications.

In addition to price reductions, Alibaba Tongyi has also launched various models of different scales to meet the needs of different scenarios. Eight large language models ranging from 500 million to 110 billion parameters cover various application scenarios from mobile to enterprise level. In addition, Tongyi has also open-sourced multimodal models such as visual understanding, audio understanding, and code generation, providing comprehensive support for developers.

Overall, major model vendors are paving the way for AI application development through a dual approach of price reduction and open sourcing. With continuous technological advancements and ongoing cost reductions, we have reason to expect that AI applications will flourish in more fields.