DeepSeek Announces Significant Price Hike for AI API Services
DeepSeek, a Hangzhou-based artificial intelligence organisation, announced plans for a "significant" price hike across its application programming interface services. This decision follows a notable increase in global demand for its new ultra-cheap AI model. The company stated specific details regarding the price adjustments would be advised to users later.
Overall pricing for these API services would be raised "in the near future", the organisation noted in a statement on its developer platform. A "significant increase" was expected, DeepSeek added, advising users to plan their usage accordingly. This move presents challenges for the Chinese start-up in maintaining its aggressively low prices amidst intense market competition.
And this shift could potentially alter its standing as a provider of cost-effective artificial intelligence solutions. An AI developer, Michael Guo, questioned the timing of the announcement in a social media post. He suggested new American models from Big Tech firms such as Meta's Muse Spark and OpenAI's GPT-5.6 Luna were competitive with DeepSeek's offerings in both capabilities and price.
DeepSeek's latest model, DeepSeek-V4-Flash-0731, a 284-billion-parameter lightweight version, had quickly gained attention within the global AI community. This model combined strong capabilities with notably low pricing. Research firm Epoch AI reported that DeepSeek-V4-Flash-0731 was the second strongest open-weight AI model on the market. It ranked behind only the massive 2.8-trillion-parameter Kimi K3 model from Beijing-based Moonshot AI. Epoch AI also placed it "roughly midway" between US lab Anthropic's Claude Opus 4.5 and Opus 4.6 models.
The current cost of DeepSeek-V4-Flash-0731 is 3 US cents per task on average, according to benchmark firm Artificial Analysis. This makes it approximately 105 times cheaper than Anthropic's Claude Fable 5, which costs USD 3.15 per task. Epoch AI further observed that DeepSeek's latest model was significantly cheaper than its Chinese open-weight counterparts.
So, the price of 14 US cents per million input tokens and 28 US cents per million output tokens was described as an "order of magnitude cheaper" than Zhipu AI's GLM-5.2 first-party API. Zhipu AI's model costs USD 1.40 per million input tokens and USD 4.40 for output. DeepSeek-V4-Flash-0731 became the "fastest growing model ever" in terms of token usage on Ollama, a Palo Alto, California-based platform that facilitates running AI models on local hardware. Ollama announced it was expanding capacity for the model in the US and Europe to meet this demand.
The ability of the Chinese start-up to offer such competitive pricing stems from both architectural optimisation and an efficient training strategy. This analysis was provided by Zhenhui Jack Jiang, a professor of innovation and information management at the University of Hong Kong's Business School. Its mixture-of-experts design, for example, activates only a small subset of experts for each input. This approach allows the model to maintain substantial capacity while using less computation and memory during inference.
DeepSeek's training method of multi-teacher knowledge distillation also contributes to its efficiency. This method combines the capabilities of several specialised models into one unified model, allowing it to absorb useful capabilities more efficiently. According to SCMP, the pending price increase reflects the pressure the company faces as its low-cost models gain widespread adoption.
DeepSeek announced a forthcoming "significant" price increase for its AI API services.
This decision follows a global surge in demand for its DeepSeek-V4-Flash-0731 model.
The model is noted for its strong capabilities and current low price point compared to competitors.
Industry experts attribute the low cost to architectural optimisation and efficient training methods.
Developer reactions suggest the price hike timing could be challenging given market competition.
Source: SCMP
