[LingXuanYin]

Summary

Qwen TTS has one model id, qwen-audio-3.1-tts-flash. 0.2.6 also accepted qwen-audio-3.0-tts-plus when a platform declaration named it; this PR takes that back out, so 0.2.7 retires the id the way #56 and #81 retired theirs.

 openai.audioSpeech validation
-  QWEN_MODELS = { qwen-audio-3.1-tts-flash, qwen-audio-3.0-tts-plus }   (0.2.6, 019aa08)
+  qwen-audio-3.1-tts-flash only                                         (as in 0.2.0-0.2.5)
 README, Text to speech
-  lists the removed ids and where they went
+  names qwen-audio-3.1-tts-flash as the Qwen TTS model
 version
-  0.2.6
+  0.2.7

Kept from 0.2.6: 56067a0 (Qwen instructions and seed parameters, role: reference_audio on the audio input) and 52e2e42.

Commits: 773bd9d reverts 019aa08; 21e73f9 bumps the version and rewrites the README line.

Evidence

• Before (0.2.6): a platform declaration named qwen-audio-3.0-tts-plus passes validateAudioSpeechRequest and is posted with that id (019aa08's tests).
  After: the revert removes those tests and the id from the adapter; such a declaration gets Unsupported audio speech model, as on 0.2.5.
• tsc -p tsconfig.json --noEmit: exit 0. vitest run: 19 files, 469 tests passed. biome check . --line-ending=crlf: 72 files, no fixes. tsdown: exit 0.
• Retired ids left in the tree:

  $ git grep -n -E 'qwen-tts|qwen-audio-3\.0|qwen-audio-3\.1-tts-plus'
  (no output, exit 1)
  

兼容性

• 退役：qwen-audio-3.0-tts-plus（0.2.6 里临时收过）在 0.2.7 拒收，和 0.2.0–0.2.5 一样。qwen-tts、qwen-audio-3.0-tts-flash、qwen-audio-3.1-tts-plus 0.2.0 / 0.2.1 起就不收，这里不变。
• 今天经 Cohub 的请求：prod 目录里语音模型只有 qwen-audio-3.0-tts-plus，qwen-tts 和 qwen-audio-3.0-tts-flash 回 generation_model_not_found（drama3-launch 2026-10-28 14:53–14:55Z 实测）。所以退役这两个，经 Cohub 的请求没有一个从成功变成失败。
• qwen-audio-3.0-tts-plus 今天能用；钉 0.2.7 的 Cohub 部署之后，用它的请求会被拒。下游（Voice Studio、Cohub ember 和 cohub-generate skill、Studio speech skill）都在这一批里改成按能力从目录挑模型。

Merge Danger

Door: two-way for the code; a published version stays.

Blast Radius: callers that name qwen-audio-3.0-tts-plus, once they upgrade to 0.2.7.

• Migration: rename the declaration or request to qwen-audio-3.1-tts-flash and keep its content and meta (one text block; voice_prompt design or one HTTP(S) reference to clone). The input needs at least 15 Unicode code points.
• Precondition for Cohub prod: deploy a Cohub that pins 0.2.7 only after Voice Studio stops calling qwen-audio-3.0-tts-plus (planned around 18:30Z) and the platform declaration is switched to qwen-audio-3.1-tts-flash.
• ^0.2.x ranges pick up 0.2.7 on their next install; Cohub pins exact versions.