iFlytek specializes in speech, language, and industry AI systems, with model services across enterprise and public-sector use cases.
11 provider cards are available in this group.
New flagship MoE released September 7, 2026: 293B total / ~30B active parameters, 200+ languages, major coding and agent upgrades, trained and served fully on domestic compute (API only).
1.7B on-device dense LLM open-sourced (Apache 2.0) on September 1, 2026 alongside the 4B edge model: up to 1M-token native context via hybrid attention, 200+ languages, trained on Huawei Ascend clusters.
Open-weights on-device edge model released September 1, 2026 by Ciyuan Spark; among the first edge models with native 1M-token context (Apache 2.0).
Spark X1 is an text generation model from Iflytek, designed for general AI application workflows.
Spark Tiny is an text generation model from Iflytek, designed for general AI application workflows.
Spark Medical Model Lite is an text generation model from Iflytek, designed for general AI application workflows.
Spark Mini is an text generation model from Iflytek, designed for general AI application workflows.
Spark Lite Patch is an text generation model from Iflytek, designed for general AI application workflows.
Spark Max is an text generation model from Iflytek, designed for general AI application workflows.
Spark Mini Instruct is an text generation model from Iflytek, designed for general AI application workflows.
Deep-reasoning flagship released February 11, 2026, trained entirely on Huawei Ascend compute; ~50% capability uplift over X1.5 with an API-controllable dynamic thinking mode (API only).