The Xunfei Spark X1 deep reasoning large model has undergone a significant upgrade, becoming the industry's only deep reasoning model trained entirely on domestic computing power. This upgrade has achieved remarkable improvements across multiple general tasks such as mathematics, coding, logical reasoning, text generation, language comprehension, and knowledge-based Q&A. Despite the model parameters being an order of magnitude smaller than similar models, its overall performance can now rival international leading standards, showcasing the strength of domestic computing power in the field of fully autonomous large models.
Xunfei Spark X1 has also made important progress in model generalization. By incorporating more complex scenario-based data, the model demonstrates industry-leading capabilities across various industrial tasks, particularly in key sectors like education, healthcare, and judiciary, further expanding its competitive edge.
On the technical front, Xunfei Spark X1 is the first to introduce a unified model of fast thinking and slow thinking, allowing a single model to support both modes simultaneously, greatly simplifying the private deployment process. Additionally, the newly upgraded model customization optimization toolchain supports both SFT and reinforcement learning optimization methods, lowering the threshold for customizations.
In terms of core technological innovation, Xunfei Spark X1 adopts a large-scale multi-stage reinforcement learning training method. Based on question complexity, it enhances the model's performance and generalization across complex reasoning, mathematics, coding, and language comprehension scenarios. Furthermore, the unified training method based on fast and slow thinking, along with innovations in engineering technology systems, ensures efficient and stable training on domestic computing platforms.
At the application level, the API services and model fine-tuning services of Xunfei Spark X1 are now fully available, empowering developers and enterprises. Developers can seamlessly integrate the Xunfei Spark X1 deep reasoning model without any barriers and enjoy limited-time free services during specific periods. Meanwhile, the Stars MaaS platform introduces the "Xunfei Spark X1 Fine-Tuning Service" for the first time, catering to diverse business needs of developers and corporate clients.
In terms of specific capability performance, Xunfei Spark X1 excels in language comprehension, text generation, mathematical problem-solving, coding abilities, and multimodal reasoning. For instance, in language comprehension, the model accurately understands complex semantics and identifies internet memes; in text generation, it can mimic classical Chinese prose styles; in math problem-solving, it addresses complex logical reasoning problems; in coding, it quickly generates user-demanded code; in multimodal reasoning, it performs deep reasoning and Q&A on images, audio, and video content.
Moreover, powered by the new X1 deep reasoning model, the Spark App officially launches a minimalist version, supporting deep reasoning and web search, offering an immersive "problem-solving" experience. Users can switch freely between the minimalist and professional versions according to their needs, enjoying a comprehensive functional experience.
This upgrade of Xunfei Spark X1 not only showcases the latest breakthroughs in domestic computing power within the field of deep reasoning large models but also provides robust technical support and rich application scenarios for a wide range of developers and enterprises.
