Physix Frontier badgePhysix Frontier
PublishChinese | EN

Physix Frontier · News Briefing Card (IT Home · Sep 23, 2026)

iFlytek Releases Spark-ASR-2.0 Speech Recognition Model

KEY FACTS

  • iFlytek officially released its Spark-ASR-2.0 large speech recognition model on September 23.
  • The model shows clear improvements in recognizing speech in complex scenarios such as dialects, high-noise environments, and low-volume audio.
  • Spark-ASR-2.0's overall inference cost is only 10% higher than the previous generation.
  • The model will be gradually rolled out to the iFlytek Input app starting September 24.
  • The iFlytek Open Platform will provide API services for Spark-ASR-2.0.

KEY DATA

10%Inference cost increase

PHYSIX OBSERVATION

Competition in speech recognition has shifted from chasing accuracy to chasing transcription quality and scenario coverage. iFlytek is trading just a 10% cost increase for an edge in complex scenarios, paving the way for deployment across input methods, office devices, glasses, and other endpoints. For users, improved usability in dialect and high-noise environments matters more in practice than benchmark scores.

Source: IT Home report