Core Insights - The launch of the deep reasoning model, iFlytek Spark X1.5, marks a significant advancement in AI technology, emphasizing the importance of achieving AI dividends through key technological breakthroughs [1][2] - iFlytek's chairman, Liu Qingfeng, highlighted four critical areas for realizing AI dividends: autonomous control, integrated hardware and software, industry depth, and personalization [1] Group 1: Product Launch and Features - iFlytek Spark X1.5 features a MoE architecture with a total parameter count of 293 billion and an activation of 30 billion, achieving a 100% improvement in inference efficiency compared to its predecessor [1] - The model has improved the efficiency of long reasoning chain reinforcement learning training from 30% to over 84%, addressing two major technical challenges [1] - iFlytek has introduced a non-autoregressive speech model architecture, which allows for parallel output of entire text sequences, resulting in a 16% performance improvement and a 520% reduction in inference costs compared to autoregressive models of the same size [1] Group 2: Integrated Solutions - iFlytek showcased a series of integrated hardware and software solutions, including the iFlytek Smart Office Book X5, which utilizes an innovative 8-microphone array to achieve a recognition accuracy of 95.08% in noisy environments [2] - The iFlytek AI translation earphones feature a unique single-ear three-microphone noise reduction system, achieving a recognition accuracy of 97.1% in complex noise conditions [2] - The iFlytek Dual-Screen Translation Machine 2.0 incorporates a strong noise reduction mode, achieving a recognition rate of 98.69% in environments with 90 dB noise [2] Group 3: Company Vision - Liu Qingfeng emphasized that AI should empower individuals to better integrate into the world and create a warmer future, reflecting iFlytek's mission to develop AI that understands and enhances human capabilities [3]

11月6日，在第八届世界声博会暨2025科大讯飞全球1024开发者节上，科大讯飞发布了基于全国产算力 的深度推理大模型讯飞"星火X1.5"。 科大讯飞董事长刘庆峰表示，蓬勃发展的AI正在深刻改变世界，AI产业红利兑现需聚焦软硬一体、行 业纵深、个性化等多个关键核心问题。目前，科大讯飞已取得显著成效，"AI红利已到兑现时刻"。 事实上，在国产算力底座能力仍较为有限的现实下，实现同样的模型训练效率和性能需要投入更长的时 间和更大的成本。 去年上半年及今年上半年，科大讯飞业绩均亏损。直至今年第三季度，科大讯飞单季度净利润转正。其 财报显示，前三季度，科大讯飞实现营收169.89亿元，同比增长14.41%；归母净利润亏损6667.54万 元，扣非后归母净利润亏损3.38亿元。第三季度，科大讯飞实现营收60.78亿元，同比增长10.02%；归 母净利润1.72亿元，同比增长202.40%；扣非后归母净利润2623.89万元，同比增长76.50%，公司经营性 现金流净额8.95亿元，同比增长25.19%。 在此次1024开发者节活动上，刘庆峰宣布，科大讯飞在全国产算力平台上，攻克了长思维链强化学习训 练效率难题，使得深度 ...