iFLYTEK to Release Latest Cognitive and Medical Large Models


iFLYTEK released the Spark X2.5 large model, which adopts a MoE architecture with parameter scales of 293B-A30B. It focuses on improving code and intelligent agent capabilities, as well as enhancing general capabilities such as mathematics and understanding. The model was trained and inferred using domestic computing power, optimizing the challenges of domestic ultra-long sequence training and inference. It has been launched on the iFLYTEK Open Platform. The edge-side 4B and 1.7B models have been open-sourced first, supporting a maximum context of 1 million Tokens.
On September 1st, iFLYTEK will open-source two endpoint general large models, Spark X2.5-4B and Spark X2.5-1.7B. They support a native 1M token long context, with enhanced capabilities in intelligent agent interaction, mathematical computation, and general understanding. These models are designed for endpoint scenarios such as automotive, smart hardware, and the Internet of Everything. On September 7th, iFLYTEK will also release a larger-scale model.
iFLYTEK revealed at its 2026 Semi-Annual Performance Briefing that it will officially launch a new main general-purpose large model based on domestically-produced computing power during this year's 1024 Developer Festival; a phased version will be released in late August for early market and developer experience, with an established technical implementation schedule.
iFlytek's AI glasses, under the iFLYTEK brand, are now available for pre-order on JD.com at a price of 4,299 yuan, with a subsidy reducing it to as low as 3,369 yuan, and optional prescription lenses. Weighing 40 grams, they feature a conventional eyewear design, optimized structure using extensive head model data, and comfort certification. The lenses are made of fully bonded resin, emphasizing AI assistant functions driven by large models.....
The competition in AI power computing has entered the deep waters of domestication. iFLYTEK released the Spark Multimodal Large Model X2-VL at the Wuxi Conference, which is currently the only model trained on fully domestic power computing. It features a specially designed architecture, marking an important breakthrough in technological iteration and self-reliance.