On May 21, Alibaba Cloud launched a blockbuster: Qwen Long, the main GPT-4 model of Tongyi Qianwen, the API input price dropped from 0.02 yuan/thousand tokens to 0.0005 yuan/thousand tokens, a direct drop of 97%.
On May 21, Alibaba Cloud launched a blockbuster: Qwen Long, the main GPT-4 model of Tongyi Qianwen, the API input price dropped from 0.02 yuan/thousand tokens to 0.0005 yuan/thousand tokens, a direct drop of 97%. This means that one yuan can buy 2 million tokens, which is equivalent to the text volume of five Xinhua Dictionaries. This model supports up to 10 million tokens of long text input, which is about 1/400 of the GPT-4 price after price reduction, breaking through the global floor price.
Qwen Long is an enhanced version of the long text model with the meaning of Qianwen. Its performance benchmarking is GPT-4, and the context length is up to 10 million. In addition to the input price dropping to 0.0005 yuan/thousand tokens, the output price of Qwen Long has also dropped by 90% to 0.002 yuan/thousand tokens. In contrast, the input prices of GPT-4, Gemini1.5 Pro, Claude 3 Sonnet and Ernie-4.0 per thousand tokens from domestic and foreign manufacturers are 0.22 yuan, 0.025 yuan, 0.022 yuan and 0.12 yuan respectively, which are much higher than Qwen long.
This price reduction of Tongyi Qianwen covers 9 commercial and open source series models. The API input price of Qwen Max, the flagship model of Tongyi Qianwen recently released, dropped to 0.04 yuan/thousand tokens, a 67% drop. Qwen Max is currently the best performing Chinese model in the industry. Its performance on the authoritative benchmark OpenCompass is equal to that of GPT-4-Turbo, and it ranks among the top 15 in the world in the Chatbot Arena.
Not long ago, Sam Altman of OpenAI forwarded the Chatbot Arena list to confirm the ability of GPT-4o. Among the top 20 models in the world, only three Chinese models were produced by Tongyi Qianwen.
The industry generally believes that AI application innovation is entering an intensive exploration period with the gradual improvement of large model performance, but the high reasoning cost is still the key factor restricting large-scale application of large models.
At the AI Smart Leaders Summit in Wuhan, Liu Weiguang, senior vice president of Alibaba Cloud Intelligent Group and president of the Public Cloud Business Department, said, "As China's largest cloud computing company, Alibaba Cloud has significantly reduced the price of big model reasoning this time, hoping to accelerate the explosion of AI applications. We expect that the number of calls to big model APIs will grow thousands of times in the future."
Liu Weiguang believes that whether it is an open source model or a commercial model, public cloud+API will become the mainstream way for enterprises to use large models, mainly for three reasons:
First, the technology dividend and scale effect of public cloud bring huge cost and performance advantages. Alibaba Cloud can continuously optimize the model itself and AI infrastructure to pursue the ultimate reasoning cost and performance. Alibaba Cloud has built an extremely resilient AI computing power scheduling system based on its self-developed core technologies and products, such as heterogeneous chip interconnection, high-performance network HPN7.0, high-performance storage CPFS, artificial intelligence platform PAI, and combined with the Bailian distributed reasoning acceleration engine, which has significantly reduced the cost of model reasoning and accelerated the speed of model reasoning.
Even with the same open source model, the call price on the public cloud is far lower than the private deployment. Taking the Qwen-72B open source model and the monthly consumption of 100 million tokens as an example, it only costs 600 yuan per month to directly call API on Alibaba Cloud Bailian, and the cost of privatization deployment exceeds 10000 yuan per month on average.
Second, it is more convenient to call multiple models on the cloud and provide enterprise level data security. Alibaba Cloud can provide each enterprise with a dedicated VPC environment to achieve computing isolation, storage isolation, network isolation, and data encryption to fully ensure data security. At present, Alibaba Cloud has led or deeply participated in the development of more than 10 major model security related international and domestic technical standards.
The third is the natural openness of cloud manufacturers, which can provide developers with the most abundant models and tool chains. Alibaba Cloud Bailian platform gathers hundreds of high-quality models at home and abroad, including Tongyi, Baichuan, ChatGLM, and Llama series. It has a built-in tool chain for big model customization and application development. Developers can easily test and compare different models, develop exclusive big models, and easily build applications such as RAG. From model selection, model adjustment, application to external service, one-stop solution.
