generated: '2026-07-21'
method: searched
source: https://maas.unisound.com/docs/release/notes
scheme: model-releases
notes: 'Token Hub publishes a 模型发布 (model release) feed rather than a dated API changelog;
  entries announce model launches and capability updates without published dates on
  the index page. The platform carries model filing number Beijing-ShanHai-20231205.'
entries:
- name: Unisound U2
  kind: text
  highlights: Agent-native LLM with autonomous planning, tool calling, multi-agent
    coordination, long-horizon task orchestration, and self-correction.
- name: U2-Med
  kind: text
  highlights: Medical/insurance/pharma LLM covering clinical, review, risk-control,
    and health-management scenarios.
- name: U2-ASR
  kind: speech
  highlights: Long-audio structured transcription; claimed industry-first 90%+ accuracy
    in complex noise and dialect scenarios; 26 languages and Chinese dialects.
- name: U2-TTS
  kind: speech
  highlights: Emotionally expressive, highly human-like text-to-speech.
- name: U2-TTS-Clone
  kind: speech
  highlights: One-sentence voice cloning with emotion transfer and Chinese-English
    cross-lingual synthesis.
- name: U1-OCR
  kind: vision
  highlights: Document classification, certificate recognition, layout restoration,
    and key-information extraction for multi-language, handwritten, and non-standard
    documents.
- name: U1-OCR-Med
  kind: vision
  highlights: Medical document parsing for records, prescriptions, reports, and billing
    documents, handling handwriting, stamps, and complex layouts.
upcoming:
- name: U2-ASR-Med
  status: announced (敬请期待)
- name: U2-TTS-Med
  status: announced (敬请期待)
- name: U2-RadiMed
  status: announced (敬请期待)