DeepSeek R2 is a next-generation generative model planned by Chinese AI company DeepSeek (DeepSeek). The original release time has not been officially disclosed, but according to media reports such as the Financial Times and Reuters, the launch process of the model may be delayed due to factors such as hardware adaptation, performance optimization and data preparation, and there is no official release timeline yet.
1. Possible reasons for the delay of DeepSeek R2 rumors
- Domestic chip training challenge
- According to the Financial Times, citing people familiar with the matter, DeepSeek had planned to use Huawei's Ascend chip for training throughout the process, but encountered problems with stability and compatibility, resulting in multiple interruptions in the training process. The report pointed out that the team eventually used Nvidia chips to complete training instead, using Ascend only in the inference stage.
- Performance optimization requirements According
- to Reuters, DeepSeek founder Liang Wenfeng is not entirely satisfied with the current performance of R2 and asked the team to further optimize it, which may have extended the development cycle.
- Data annotation progress slows down Industry
- media said that the large-scale annotation data required for R2 was complex and exceeded the expected timeline.
2. Release time and industry expected
- release time window According
- to industry sources, the R2 may be released in mid-to-late August 2025, and the hardware architecture may use the Ascend 910B chip cluster, with performance claimed to be close to 91% of the Nvidia A100 cluster.
- Cautious market attitude Some
- analysts believe that the current claims about the release time are still rumory and recommend waiting for official confirmation.
3. Industry significance and potential impact
- R2 is considered an important upgrade after R1 and V3, which is expected to improve multilingual reasoning capabilities and optimize inference costs.
- The test of the domestic computing power ecology
- Some analysts pointed out that this delay may reflect the shortcomings of domestic AI chips in large-scale model training, and also expose the challenges in the promotion of localization strategies.
FAQ
Q: Has R2 been confirmed to be postponed?
A: The official news of the postponement has not been confirmed, and the relevant information mainly comes from media and industry reports.
Q: What competitors does the R2 have?
A: Including the Alibaba Qwen series and other next-generation products from domestic and foreign large model manufacturers.
Q: Will it be completely domestic chips?
A: According to reports, the training phase mainly relies on Nvidia chips, and domestic chips are currently mostly used for inference.
