China's DeepSeek is ready with version 2. What to expect

First Post

29-04-2025

Chinese media reports claim that DeepSeek R2 will be 97.3 per cent cheaper than OpenAI's GPT-4o model. The new version of the artificial intelligence model will completely rely on Huawei's Ascend 910B GPU cluster, signalling a total independence from American-made AI chips read more
Globally, DeepSeek has faced increasing scrutiny since its release in January. Although the tool gained initial acclaim for its efficient open-source AI model, governments have raised red flags over its data practices. Image Credit: Reuters
DeepSeek, the Chinese tech giant that turned around the US stock market a couple of months ago, is ready with an advanced model to be released this week. DeepSeek R2 will reportedly be cheaper and better, giving tough competition to ChatGPT's maker OpenAI.
Chinese media reports claim that DeepSeek R2 will be 97.3 per cent cheaper than OpenAI's GPT-4o model. The new version of the artificial intelligence model will completely rely on Huawei's Ascend 910B GPU cluster, signalling a total independence from American-made AI chips.
STORY CONTINUES BELOW THIS AD
How will it be different?
Market analysts noted on X that if the new DeepSeek model matches rival performance on global benchmarks, DeepSeek could position Huawei as the first major challenger to NVIDIA. The earlier model released by the Hangzhou-based AI startup had triggered a massive sell-off in US stock markets, erasing $1.5 trillion in value and causing sharp declines in the shares of tech giants such as NVIDIA, Alphabet, and Microsoft.
DeepSeek is also planning to dethrone Meta in dominating the open-source AI category by making its own models free to use. This comes ahead of OpenAI's plan to release its first open-source model soon.
DeepSeek R2 is expected to be similar in scale to OpenAI's largest model to date, GPT-4.5 (code-named Orion), which has 1.8 trillion parameters. In comparison, DeepSeek R2 is projected to feature 1.2 trillion parameters and will be trained on 5.2 petabytes of data.
Alibaba joins the AI game
China is readying itself to join the AI race. Earlier today, Tech giant Alibaba Group has joined the AI race after launching its artificial intelligence model Qwen 3, an upgraded version of its flagship model that is equipped with new hybrid reasoning capabilities.
The Qwen3 series features two mixture-of-experts (MoE) models designed to compete with hybrid reasoning systems, recently launched by Anthropic and Google, a subsidiary of Alphabet.
Earlier this year, Alibaba announced its full commitment to the AI race and, just weeks ago, released a new model from its Qwen 2.5 series capable of processing text, images, audio, and video. The model is efficient enough to run on smartphones and laptops, and in March, the company also introduced an updated version of its AI assistant, the Quark app.