Chinese artificial intelligence startup DeepSeek released a preview version of its new model V4 on Friday.
The long-awaited next-generation model “includes an ultra-long reference of one million words,” the company said in a statement on social media.
DeepSeek-V4 is available in a Pro version and a cheaper Flash version. V4-Pro has 1.6 trillion parameters while V4-Flash has 284 billion parameters, which determine the decision-making ability of the model.
“In global knowledge benchmarks, DeepSeek-v4-Pro significantly outperforms other open-source models and performs only slightly better than the top-tier closed-source model, (Google’s) Gemini-Pro-3.1,” the company’s statement said.
Releasing a preview version allows the company to incorporate real-world feedback before finalizing the model.
US-China AI race
DeepSeek caused a stir in January last year when it unveiled a generative AI chatbot with capabilities that rival US products like ChatGPT, but it said it took significantly less computing power and money to develop.
The company has also been in controversies. For example, its chatbot has been found to avoid questions on politically sensitive topics such as the 1989 Tiananmen crackdown, raising questions about censorship.
The Hangzhou-based startup has also been accused of unfair and illegal conduct by the United States and its American competitors.
On Thursday, the White House alleged that Chinese entities were engaging in an “industrial-scale distillation campaign to steal American AI.”
Beijing rejected the “groundless allegations”, and said China “attaches great importance to the protection of intellectual property rights.”
Edited by: Shawn Sinico