China’s DeepSeek has released its latest AI model, V4, as part of its push to compete with leading systems from US companies. The Hangzhou-based firm said its new open-source model is designed to match the performance of closed-source AI models developed by companies such as OpenAI and Google DeepMind. The launch includes two versions of the model – DeepSeek-V4-Pro with 1.6T parameters and DeepSeek-V4-Flash with 284B parameters. The release, marking one of the company’s largest developments so far, comes as competition in the global AI market continues to grow, with companies focusing on scale, performance and cost efficiency.“In world knowledge benchmarks, DeepSeek-V4-Pro significantly leads other open-source models and is only slightly outperformed by the top-tier closed-source model, (Google’s) Gemini-Pro-3.1,” the company said in a statement.
DeepSeek previews V4 AI model
As mentioned above, DeepSeek released two versions of the model: V4-pro and V4-flash. The V4-pro model has 1.6 trillion parameters, making it the company’s largest model to date. The smaller V4-flash model has 284 billion parameters.Both versions support a context window of 1 million tokens, which determines how much information the system can process at one time. The company said this was achieved with high cost efficiency.“Through architectural innovations, DeepSeek-V4 series achieve a dramatic leap in computational efficiency for processing ultra-long sequences. This breakthrough enables efficient support for a context length of one million tokens, ushering in a new era of million-length contexts for next-generation LLMs,” the company says. “We believe our ability to efficiently handle ultra-long sequences unlocks the next frontier of test-time scaling, paves the way for deeper research into long-horizon tasks, and establishes a necessary foundation for exploring future paradigms like online learning,” it added.
DeepSeek V4 AI model ‘s hardware and development details
DeepSeek did not disclose the exact hardware used to train the V4 models. However, it said its system includes software components designed to work with both Nvidia and Huawei chips.The company noted that performance is currently limited by available computing capacity. It added that costs are expected to decrease later in the year as new hardware, including Huawei’s Ascend 950PR systems, becomes available at scale.The release comes amid ongoing restrictions on advanced semiconductor exports to China, particularly high-end graphics processing units from Nvidia. These restrictions have affected the development of AI models in the country.
