iFlytek Input Method 16.0 turns the keyboard into a drafting assistant
The AMW Read
An incremental consumer product update from a known Chinese model maker: it spreads Spark speech and drafting models into the keyboard, but it changes little beyond iFlytek's own consumer distribution.
iFlytek Input Method 16.0 turns the keyboard into a drafting assistant
iFlytek has released version 16.0 of its iFlytek Input Method. The main new feature is "intent-based drafting": the user states a task and supplies context, and the keyboard organizes that information into a draft for the user to review. The release also adds the Spark-ASR-2.0 speech recognition model, support for 202 dialects, and more editing and style controls, which iFlytek presents as a personalized AI keyboard.
The input method is one of the most-used surfaces on a phone, and iFlytek is turning it into a place to deploy its own Spark models. Most labs reach users through a chat app they have to open separately. Here the drafting help sits inside the text field people already use. Speech is where iFlytek has historically been strongest, and 202 supported dialects is an advantage that general chatbot competitors do not obviously match. The release also follows iFlytek subsidiary Ciyuan Xinghuo open-sourcing its on-device Spark X2.5 models with 1M-token context for smart hardware in September. Together, the two moves show iFlytek building out its Spark family through devices and everyday interfaces, not just through standalone model releases. The article does not report user numbers, how the features are split between on-device and cloud processing, or pricing.
For builders, the takeaway is that drafting features with a review step built into the keyboard put pressure on standalone writing-assistant apps. If the operating system's input layer can take a task and context and produce a draft, single-purpose writing tools lose their reason to exist. Investors should watch whether iFlytek can turn its speech and dialect coverage into lasting use of Spark-ASR-2.0 and the drafting features. Better speech recognition alone could become a commodity as rival speech models improve.


