DeepSeek, the Chinese AI startup, announced on Wednesday that it will offer discounts for its API platform during non-peak hours – from 16:30 – 00:30 every day.
During this off-peak period, the cost for DeepSeek-V3 is reduced by 50% to $0.035 per million tokens for input (cache hit), $0.135 for input (cache miss), and $0.550 for output.
Meanwhile, DeepSeek-R1 sees a 75% reduction, with prices dropping to $0.035 per million tokens for input (cache hit), $0.135 for input (cache miss), and $0.550 for output.
In comparison, OpenAI’s o1 reasoning model charges $60 for output and $15 for 1 million input API tokens.
That said, DeepSeek’s API recently encountered a series of problems. At one point, its status indicated that it experienced downtime for 10 continuous days. On February 6, it was also reported that DeepSeek temporarily suspended API top-ups for developers.
DeepSeek to Release R2 Model Before May
On Tuesday, Reuters reported that DeepSeek plans to release its next reasoning model, the DeepSeek R2, ‘as early as possible’. The company initially planned to release it in early May but is now considering an earlier timeline.
The model is said to produce ‘better coding’ and reason in languages beyond English.
The competition in China’s AI ecosystem is heating up, and recently, Alibaba introduced a preview of the Qwen QwQ-Max reasoning model, and the company also committed to a $52 billion investment in AI infrastructure over the next three years.
Besides its plans to release new models, DeepSeek is keen to open-sourcing its technologies. The startup announced an open-source week, where it will release five new open source repositories.
On Wednesday, the company announced its third release called DeepGEMM, an FP8 GEMM library optimised for dense and Mixture of Experts (MoE) computations. The library is said to deliver more than 1350 FP8 TFLOPS on NVIDIA Hopper GPUs.
Recently, the startup released its DeepSeek-R1 and DeepSeek-V3 models, creating a storm across the AI ecosystem.
These models offered state-of-the-art performance while being trained and deployed at a fraction of the cost of their competitors while also being available as open source.
