The large language model (LLM) arena is heating up again, with DeepSeek, known as the "price butcher," initiating another round of price cuts. This move raises the question: will other players follow suit, and what does this mean for the future of LLMs?
DeepSeek recently announced a significant reduction in its API input costs to 0.1 yuan per million tokens and output costs to 2 yuan per million tokens. This price slash represents another order of magnitude decrease in LLM API pricing.
Key Takeaways:
DeepSeek attributes this price reduction to its innovative use of contextual hard drive caching. The company explains that a significant portion of user input in LLM API usage is often repetitive.
How Contextual Hard Drive Caching Works:
DeepSeek is the first global LLM provider to widely adopt hard drive caching in its API service. This is made possible by the MLA (Mixture of LoRA Experts) structure introduced in DeepSeekV2, which improves model performance while significantly compressing the context KVCache size, reducing storage and bandwidth requirements, allowing for caching on low-cost hard drives.
Additional Advantages:
This isn't DeepSeek's first foray into price wars. Since May, the company has been a disruptor in API pricing.
Timeline of DeepSeek's Price Cuts:
This initial price reduction triggered a wave of responses from industry players like Zhipu AI, Volcano Engine, Baidu, Tencent, and Alibaba Cloud, who all announced price cuts. Alibaba Cloud's Tongyi Qianwen core model Qwen-Long saw a staggering 97% price decrease, landing at 0.0005 yuan/thousand tokens. Baidu and Tencent even offered some models for free. Internationally, OpenAI's GPT-4o was released with free usage and halved API call prices.
Volcano Engine's Doubao general model pro-32k was priced at only 0.0008 yuan/thousand tokens, a 99.3% reduction compared to the industry average of 0.12 yuan/thousand tokens, pushing the market into the "cent era." Volcano Engine's president, Tan Dai, stated that reducing costs is crucial for accelerating the transition to a "value creation stage."
The driving force behind these price reductions is to lower the barrier to entry for businesses to adopt and innovate with LLMs. A Volcano Engine insider noted that the lack of widespread enterprise application of LLMs necessitates lower prices to encourage adoption.
Key Objectives:
However, the financial sustainability of relying solely on API sales remains a concern. "No large model company survives by selling APIs alone," stated one FA (Financial Advisor) familiar with the LLM industry. 猎豹移动 ( Cheetah Mobile ) Chairman and CEO Fu Sheng believes that significant price cuts force LLM startups to explore new business models. Large companies with cloud services can afford to lower prices to attract more customers.
Unlike the previous wave of price cuts, DeepSeek's latest move has not yet been met with immediate responses from other major LLM companies. However, this continued downward price trend suggests that the democratization of LLMs is underway, and the vertical application ecosystem is poised for further growth.
Possible Outcomes:
The LLM landscape is evolving rapidly, and these price wars are a testament to the increasing competition and the push for wider adoption. As LLMs become more accessible, we can expect to see even more innovative applications emerge across various sectors.
