iFlytek Spark

iFlytek Spark is a series of cognitive models (including X1.5, X2) built entirely on domestic computing power with a MoE architecture. It delivers capabilities in language understanding, knowledge Q&A, logical reasoning, mathematics, and code generation, covering over 130 languages. Through a free API and agents such as Spark Teacher Assistant and Spark Industry Analyst, users can generate teaching plans, create courseware, prepare exam papers, and assist with industry analysis reports. Key strengths include a 100% leap in inference efficiency from X1 to X1.5 and further gains in X2, internationally leading math performance, and strong multilingual ability (especially in Latin America and ASEAN languages), reaching over 95% of GPT-5 performance. Features like personalized memory and voice cloning add practical value. One known drawback is that in engineering paper generation tests, outputs lacked solid literature support, contained imprecise data and occasional fabrications, making the tool unreliable for rigorous academic writing without human verification. Rapid iteration may also cause occasional stability concerns, and the ecosystem is still maturing.