At the "AI Without Limits Global Intelligence" launch conference held in Shanghai, iFlytek announced a significant upgrade to its deep inference model, Starfire X1. Yu Jidong, iFlytek's senior vice president, revealed that this update will further enhance Starfire X1's performance in inference, text generation, and language comprehension, aiming to match industry-leading models like OpenAI o1 and DeepSeek R1.
Starfire X1 was first launched in January 2025 and attracted industry attention for its training using domestically produced computing power. Utilizing 10,000 domestically produced 910B computing cards, iFlytek achieved a series of technological breakthroughs through extensive adaptation and efficiency optimization work on the national computing platform. In this update, iFlytek states that while maintaining its advantage in mathematical tasks, it will strive for even greater progress in general tasks.
Over the past few months, iFlytek has been intensively training the new version of Starfire X1. The company expects this version to be ready within the next three months and surpass or match OpenAI o1 in solving mathematical problems and chain-of-thought reasoning capabilities. With the continued advancement of domestic computing power, iFlytek's efforts have laid a solid foundation for adapting large domestic models to domestic computing power.
iFlytek's announcement marks an intensification of competition in the artificial intelligence field. Several domestic and international companies are working on inference and text generation models, vying for market share and technological leadership. With the powerful upgrade of Starfire X1, iFlytek will further enhance its position in this competition and drive the development of artificial intelligence technology.
Key Points:
🌟 The updated version of Starfire X1 will be benchmarked against OpenAI o1 and DeepSeek R1 in inference, text generation, and language comprehension.
💡 Starfire X1 was first launched in January 2025 and achieved significant technological breakthroughs using domestically produced computing cards.
🚀 iFlytek expects the new version to be ready in three months and surpass OpenAI o1 in mathematical capabilities.
