There have been reports that the DeepSeek R2 AI model will be announced later this month. The new update is said to rival the OpenAI recently unveiled Chat GPT 5 edition. However, sources close to DeepSeek have refuted the rumors, which means the R2 model will not be announced this month.
DeepSeek R2 AI model’s arrival delayed

Around three months ago, a Reuters report claimed that the DeepSeek R2 AI model would be released in May. It was later reported at that time that the company’s CEO, Lian Wenfeng, was not satisfied with the P2’s performance, and engineers were working on refining the experience.
The arrival of the R2 AI model was reportedly further delayed due to the US imposing export restrictions on Nvidia chips that were supposed to power the AI model. However, recent reports have claimed that the R2 will be powered by Huawei’s Ascend 910B AI chips for training and development. The switch to the Huawei chip is said to be for better accessibility in China and to overcome supply issues.
Previously, while the DeepSeek R1 model used the Huawei Ascend 910C for inference, the model was trained on Nvidia GPUs (H100/H800). As far as the R2 is concerned, it is said to be not only trained but also designed on the Huawei Ascend 910B chip. As mentioned above, new rumors suggest that the R2 AI model will not be released this month, and there is no official release timeframe yet.
DeepSeek R2: What to expect?
Rumors have revealed that the DeepSeek R2 will have stronger coding capabilities as well as better reasoning across different languages beyond English. It will be a massive model equipped with 1.2 trillion parameters and trained on 5.2 petabytes of data.
The R2 is said to feature an advanced MoE (Mixture of Experts) architecture, combining a smart gating network layer to efficiently improve performance on demanding tasks. It is also expected to offer better image processing, longer context handling, and more concise outputs.
While the DeepSeek R1 is free to use, people need to pay to access its API. The same is expected from the R2 AI model. However, it is said to be 97.3 percent cheaper than OpenAI’s Chat-4o in terms of training and inference costs.
About six months ago, DeepSeek R1 drew massive attention for delivering strong reasoning skills at a fraction of GPT-4’s cost, but its buzz has since faded as limitations in multilingual reasoning and image handling emerged. Expectations are now high that R2 will address these gaps with improved language coverage, stronger coding abilities, and enhanced multimodal performance.
For more daily updates, please visit our News Section.
Stay ahead in tech! Join our Telegram community and sign up for our daily newsletter of top stories.
(via)

Comments