iFLYTEK has officially introduced the Spark X1.5 deep reasoning large model, a cutting-edge innovation that leverages entirely domestic computing resources. This model adopts the MoE (Mixture of Experts) architecture, boasting a staggering total of 293 billion parameters, of which 30 billion are actively utilized. Remarkably, it can be seamlessly deployed on a single standard Ascend server.
Compared to its predecessor, the X1 model, Spark X1.5 has achieved a remarkable 100% improvement in reasoning efficiency. Its end-to-end performance surpasses 93% of the efficiency demonstrated by international competitors. The six core capabilities of Spark X1.5 are on par with those of international mainstream large models, with its mathematical prowess standing out on a global scale.
In the Gaokao (Chinese College Entrance Examination) math assessment, it outperformed GPT-5 and other models, and it is capable of accurately solving problems from the International Mathematical Olympiad. Its performance in tasks related to education and healthcare also exceeds that of GPT-5 and its counterparts.
After enhancing its multilingual capabilities, Spark X1.5 now supports over 130 languages, with an overall performance that exceeds 95% of ChatGPT-5's capabilities, and it takes the lead in several key languages. Furthermore, the Spark Simultaneous Interpretation Large Model 2.0 has secured the top spot in the industry across eight core dimensions evaluated by IDC.