On April 21st, iFlytek officially announced its latest upgraded AI model – iFlytek StarFire X1, aiming to rival OpenAI's o1 and DeepSeek R1 in intelligent reasoning and multi-tasking. This domestically-trained deep reasoning large model, utilizing entirely national computing power, demonstrates outstanding performance across various general tasks, including mathematics, programming, logical reasoning, text generation, language understanding, and knowledge question answering.
In this upgrade, iFlytek StarFire X1 significantly enhances its generalization ability by incorporating data from more complex scenarios, resulting in superior performance in key sectors such as education, healthcare, and the judicial system. iFlytek stated that although StarFire X1's model parameters are an order of magnitude smaller than comparable industry products, its overall effectiveness is on par with leading competitors.
Notably, iFlytek StarFire X1 introduces a unified "fast thinking, slow thinking" model, allowing users to meet diverse thinking needs with a single model. This innovation simplifies private deployment and lowers the barrier to entry for businesses using AI. Businesses only need 4 Huawei 910B cards for full deployment, while industry-specific optimization requires only 16 cards.
Furthermore, iFlytek released a newly upgraded model customization and optimization toolkit, supporting both SFT and reinforcement learning model customization schemes, enabling businesses to more easily implement AI applications. The StarFire X1 API is also open, facilitating access for developers and businesses.
The significant upgrade of iFlytek StarFire X1 undoubtedly provides strong AI support across various industries, signaling the continuous improvement of domestic AI technology's global competitiveness.
