Core Insights - The company iFlytek officially launched the Xunfei Spark X1.5 deep reasoning model based on fully domestic computing power on November 6 [1] - This model has achieved over 93% efficiency compared to international competitors in end-to-end performance by overcoming the full-link training efficiency of the MoE model [1] - The Spark model supports over 130 languages, with overall performance reaching over 95% of GPT-5 [1]

人民财讯11月6日电，11月6日，科大讯飞（002230）发布基于全国产算力的讯飞星火X1.5深度推理大模 型，率先在全国产算力平台上攻克MoE模型全链路训练效率，端到端性能达到国际竞品效率的93%以 上。据悉，该模型其语言理解、文本生成、知识问答、逻辑推理、数学能力、代码能力等对标国际主流 大模型。同时，星火多语言能力持续升级，支持130+种语言，整体性能达到GPT-5的95%以上。 ...

Core Insights - On November 6, iFlytek launched the Xunfei Spark X1.5 deep reasoning model based on fully domestic computing power, achieving over 93% efficiency compared to international competitors in end-to-end performance [1] - The model excels in language understanding, text generation, knowledge Q&A, logical reasoning, mathematical capabilities, and coding skills, aligning with mainstream international large models [1] - The multilingual capabilities of the Spark model have been continuously upgraded, now supporting over 130 languages and achieving 95% overall performance compared to GPT-5 [1]