AI News (2026/2/11): iFLYTEK Spark X2 Large Model Officially Released, API Launched Simultaneously
Executive Summary:
On Wednesday, February 11, iFLYTEK officially released the Spark X2 large model, which is trained on entirely domestic computing power, and launched the API. Spark X2 matches international top-tier models in core metrics such as mathematics, reasoning, and language understanding, supports over 130 languages, and has improved reasoning performance by 50% compared to its predecessor, achieving deployment on a single domestic Ascend server.
iFLYTEK SparkX2 Large Model Officially Released, API Launched Simultaneously
News Details
On Wednesday, February 11, iFLYTEK announced the official release of its Spark X2 large model, which is trained on entirely domestic computing power, and the simultaneous launch of the API interface. Spark X2 has achieved international top-tier levels in multiple core metrics, particularly in mathematics, reasoning, and language understanding. The model supports over 130 languages, further solidifying its leading position in multilingual processing.
Key Points
【293B MoE Sparse Architecture】: Spark X2 employs a 293B parameter MoE (Mixture of Experts) sparse architecture, significantly enhancing the model's reasoning performance and resource utilization through an efficient expert combination mechanism.
【50% Improvement in Reasoning Performance】: Compared to its predecessor, Spark X1.5, Spark X2's reasoning performance has improved by 50%. This enhancement is primarily due to multiple engineering optimizations and technical improvements, making the model more efficient and reliable in practical applications.
【Deployment on a Single Domestic Ascend Server】: Spark X2 has achieved deployment on a single domestic Ascend server, reducing deployment costs and complexity. This not only helps promote the application of domestic computing power but also provides more options for small and medium-sized enterprises.
AI-ALL In-Depth Commentary
The release of iFLYTEK's Spark X2 large model marks a further maturation of domestic AI technology in the large model domain. By adopting a 293B parameter MoE sparse architecture, Spark X2 not only matches international top-tier levels in performance but also excels in multilingual support. The significant improvement in reasoning performance and the capability to deploy on a single Ascend server make the model more competitive in practical applications. For AI developers and enterprises, the launch of the Spark X2 API provides a powerful tool option, accelerating the implementation and application of AI technology. Additionally, this progress highlights the rise of domestic computing platforms in the high-performance computing field, injecting new momentum into the development of the domestic AI ecosystem.
