iFLYTEK Spark

iFLYTEK Spark (讯飞星火) is a cognitive large language model independently developed by iFLYTEK, positioned as an artificial general intelligence (AGI) platform. Trained on a fully domestic computing stack, it integrates core capabilities including text generation, language understanding, knowledge Q&A, logical reasoning, mathematical computation, code generation, and multimodal interaction. As one of China's first large models trained on Huawei Ascend AI chips and the "Feixing-1" computing platform, Spark stands out in Chinese understanding, voice interaction, and mathematical reasoning.

• Multimodal agent platform: 63,000+ public agents covering all scenarios—year-end summaries, PPT creation, code writing, travel planning, and more
• Ultra-realistic voice interaction: Powered by Spark's ultra-realistic low-latency interaction technology, with multiple emotions, styles, and dialects, free interruptions, and extremely fast response
• Deep reasoning: Strong autonomous thinking and planning, calling on web search, a code interpreter, image understanding, and other tools to handle complex tasks
• Multi-format document processing: Upload Word, PPT, PDF, Excel, and other formats with quick-read parsing, summarization, polishing, and rewriting
• AI search and creation: Aggregates authoritative content in real time, precisely distills and integrates results, and supports all types of writing—content creation, video scripts, marketing copy, and more
• Image and video generation: Generates images in styles from cinematic portraits and children's drawings to cyberpunk, with customizable virtual human videos

In enterprise applications, Spark deeply empowers education, healthcare, finance, government, and other industries; for personal use, the app delivers multimodal interaction, intelligent assistants, and content creation; in the developer ecosystem, complete APIs and SDKs support enterprise application integration.

iFLYTEK Spark V4.0 topped MIT Technology Review's evaluation as the "smartest" domestic large model with a score of 81.5, ranking first in math ability, science comprehensive, and coding assessments. As a fully self-controlled AI large model with a domestically developed stack, it not only has strong technical capability but also provides important support for China's domestic AI industry.