feat!: move Qwen TTS to the 3.1 ids (flash, plus); drop qwen-tts and the 3.0 ids

README.md

@@ -198,7 +198,7 @@ Agents and external tools should inspect a model declaration before constructing
 import { createGenerationClient } from "@neta-art/generation";
 
 const discoveryClient = createGenerationClient();
-const declaration = discoveryClient.getModel("qwen-tts");
+const declaration = discoveryClient.getModel("qwen-audio-3.1-tts-flash");
 if (!declaration) throw new Error("Model is unavailable");
 
 console.log(discoveryClient.stringifyModelConfig(declaration.model, { format: "json" }));
@@ -220,7 +220,7 @@ The same declarations can be exported as YAML through the existing CLI:
 
 ```bash
 neta-generation models list
-neta-generation models export qwen-tts --out ./qwen-tts.yaml
+neta-generation models export qwen-audio-3.1-tts-flash --out ./qwen-audio-3.1-tts-flash.yaml
 neta-generation models export-all --out ./models
 ```
 
@@ -308,9 +308,8 @@ const client = createGenerationClient({
 - `gpt-image-2`
 - `z-image-turbo`
 - `qwen-image-edit`
-- `qwen-tts`
-- `qwen-audio-3.0-tts-plus`
-- `qwen-audio-3.0-tts-flash`
+- `qwen-audio-3.1-tts-flash`
+- `qwen-audio-3.1-tts-plus`
 - `gemini-3.1-flash-image-preview`
 - `kling-text-to-video`
 - `kling-image-to-video`
@@ -403,30 +402,33 @@ await client.generate({
 
 ## Text to speech
 
+`qwen-tts`, `qwen-audio-3.0-tts-plus` and `qwen-audio-3.0-tts-flash` are removed in this release. Migrate to `qwen-audio-3.1-tts-flash`, or to `qwen-audio-3.1-tts-plus`, which this release adds; both need input of at least 15 Unicode code points.
+
 Each TTS request accepts exactly one non-empty text block and returns one URL audio block. Choose the model from the requested voice mode and priority before constructing the request:
 
-| Requirement | Model choice |
+| Requirement | Model choice (default tier) |
 | --- | --- |
-| Create a voice from a text-only description, without reference audio | Use an explicitly requested Qwen variant; otherwise use `qwen-tts` as the deterministic default |
-| Clone a voice from one reference recording | Use an explicitly requested Qwen variant, or `qwen-tts`; pass one HTTP(S) reference URL and no `voice_prompt` |
-| Use a default voice, including a delegated choice expressed only as any, random, suitable, or natural | `qwen-tts` with a short generic `voice_prompt`, such as a natural, neutral narrator |
+| Create a voice from a text-only description, without reference audio | `qwen-audio-3.1-tts-flash` with a `voice_prompt` |
+| Clone a voice from one reference recording | `qwen-audio-3.1-tts-flash`; pass one HTTP(S) reference URL and no `voice_prompt` |
+| Use a default voice, including a delegated choice expressed only as any, random, suitable, or natural | `qwen-audio-3.1-tts-flash` with a short generic `voice_prompt`, such as a natural, neutral narrator |
 
-- Qwen: `voice_prompt` design OR one-reference clone; `qwen-tts` is the unspecified-design default and accepts any text length; Plus / Flash require at least 15 Unicode code points.
+- Qwen: `voice_prompt` design OR one-reference clone via `qwen-audio-3.1-tts-flash` (default) or `qwen-audio-3.1-tts-plus`; both require at least 15 Unicode code points.
+- Tier: the table names `qwen-audio-3.1-tts-flash`, the default. In any row, use `qwen-audio-3.1-tts-plus` instead only when the user or an external policy selects it. Alibaba describes Flash as its low-latency tier and Plus as its quality tier for audiobooks and dubbing; the SDK has not measured either.
 - Conflict: reference + redesign requires user choice before generation.
 - Dependency: clone prior generated audio.
-- Ranking: no declared Qwen quality, latency, or cost order.
+- Short text: input under 15 Unicode code points is rejected; ask the user to lengthen it.
 
 ```ts
 await client.generate({
-  model: "qwen-tts",
-  content: [{ type: "text", text: "欢迎使用语音合成功能。" }],
+  model: "qwen-audio-3.1-tts-flash",
+  content: [{ type: "text", text: "欢迎使用语音合成功能，这是一段示例文本。" }],
   meta: {
     voice_prompt: "一位沉稳自然的中文播音员，吐字清晰，语速适中",
   },
 });
 
 await client.generate({
-  model: "qwen-audio-3.0-tts-flash",
+  model: "qwen-audio-3.1-tts-flash",
   content: [
     { type: "text", text: "这是一段长度足够并且表达清晰自然的语音合成文本。" },
     { type: "audio", source: { type: "url", url: "https://example.com/reference.mp3" } },

examples/text-to-speech.ts

@@ -5,7 +5,7 @@ if (!apiKey) throw new Error("Set NETA_ROUTER_API_KEY or NETA_API_KEY");
 
 const client = createGenerationClient({ apiKey });
 const output = await client.generate({
-  model: "qwen-tts",
+  model: "qwen-audio-3.1-tts-flash",
   content: [{ type: "text", text: "欢迎使用语音合成功能，这是一段示例文本。" }],
   meta: {
     voice_prompt: "一位沉稳自然的中文播音员，吐字清晰，语速适中",

models/qwen-audio-3.0-tts-flash.yaml → models/qwen-audio-3.1-tts-flash.yaml

@@ -1,7 +1,8 @@
 schema: neta.generation.model.v1
-model: qwen-audio-3.0-tts-flash
-title: Qwen Audio 3.0 TTS Flash
-description: 'Modes: voice_prompt design OR one-reference clone. Text: >=15 Unicode code points. Conflict: ask user; never combine/reinterpret. Dependency: clone prior generated audio.'
+model: qwen-audio-3.1-tts-flash
+title: Qwen Audio 3.1 TTS Flash
+description: "Modes: voice_prompt design OR one-reference clone. Text: >=15 Unicode code points. Tier: default;
+  Alibaba's low-latency tier. Conflict: ask user; never combine/reinterpret. Dependency: clone prior generated audio."
 adapter:
   type: openai.audioSpeech
 content:
@@ -16,25 +17,25 @@ content:
       max: 1
       sources:
         - url
-      description: 'Clone: one URL; no voice_prompt. Dependency: use prior generated audio.'
+      description: "Clone: one URL; no voice_prompt. Dependency: use prior generated audio."
 meta:
   fields:
     voice_prompt:
       type: string
       optional: true
-      description: 'Design: custom voice text; no reference audio.'
+      description: "Design: custom voice text; no reference audio."
 examples:
   - title: Voice design
     request:
-      model: qwen-audio-3.0-tts-flash
+      model: qwen-audio-3.1-tts-flash
       content:
         - type: text
           text: 这是一段长度足够并且表达清晰自然的语音合成测试文本。
       meta:
         voice_prompt: 一位沉稳干练的男性播音员声音，吐字清晰有力
   - title: Voice clone
     request:
-      model: qwen-audio-3.0-tts-flash
+      model: qwen-audio-3.1-tts-flash
       content:
         - type: text
           text: 这是一段长度足够并且表达清晰自然的语音合成测试文本。

models/qwen-audio-3.0-tts-plus.yaml → models/qwen-audio-3.1-tts-plus.yaml

@@ -1,7 +1,9 @@
 schema: neta.generation.model.v1
-model: qwen-audio-3.0-tts-plus
-title: Qwen Audio 3.0 TTS Plus
-description: 'Modes: voice_prompt design OR one-reference clone. Text: >=15 Unicode code points. Conflict: ask user; never combine/reinterpret. Dependency: clone prior generated audio.'
+model: qwen-audio-3.1-tts-plus
+title: Qwen Audio 3.1 TTS Plus
+description: "Modes: voice_prompt design OR one-reference clone. Text: >=15 Unicode code points. Tier: Alibaba's quality
+  tier (audiobooks, dubbing); only when user or external policy selects it. Conflict: ask user; never
+  combine/reinterpret. Dependency: clone prior generated audio."
 adapter:
   type: openai.audioSpeech
 content:
@@ -16,25 +18,25 @@ content:
       max: 1
       sources:
         - url
-      description: 'Clone: one URL; no voice_prompt. Dependency: use prior generated audio.'
+      description: "Clone: one URL; no voice_prompt. Dependency: use prior generated audio."
 meta:
   fields:
     voice_prompt:
       type: string
       optional: true
-      description: 'Design: custom voice text; no reference audio.'
+      description: "Design: custom voice text; no reference audio."
 examples:
   - title: Voice design
     request:
-      model: qwen-audio-3.0-tts-plus
+      model: qwen-audio-3.1-tts-plus
       content:
         - type: text
           text: 这是一段长度足够并且表达清晰自然的语音合成测试文本。
       meta:
         voice_prompt: 一位沉稳干练的男性播音员声音，吐字清晰有力
   - title: Voice clone
     request:
-      model: qwen-audio-3.0-tts-plus
+      model: qwen-audio-3.1-tts-plus
       content:
         - type: text
           text: 这是一段长度足够并且表达清晰自然的语音合成测试文本。

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@neta-art/generation",
-  "version": "0.1.42",
+  "version": "0.2.0",
   "description": "A lightweight multimodal generation SDK with built-in model presets and adapter-based provider calls.",
   "keywords": [
     "ai",

src/adapters/audio-speech.ts

@@ -8,8 +8,7 @@ import type {
 } from "../types.js";
 
 const REQUEST_TIMEOUT_MS = 210_000;
-const QWEN_MODELS = new Set(["qwen-tts", "qwen-audio-3.0-tts-plus", "qwen-audio-3.0-tts-flash"]);
-const QWEN_AUDIO_3_MODELS = new Set(["qwen-audio-3.0-tts-plus", "qwen-audio-3.0-tts-flash"]);
+const QWEN_MODELS: ReadonlySet<string> = new Set(["qwen-audio-3.1-tts-flash", "qwen-audio-3.1-tts-plus"]);
 
 type TextBlock = Extract<GenerationContentBlock, { type: "text" }>;
 type AudioBlock = Extract<GenerationContentBlock, { type: "audio" }>;
@@ -91,7 +90,7 @@ function validateQwen(input: ResolvedGenerationRequest, text: TextBlock, audio:
     );
   }
 
-  if (QWEN_AUDIO_3_MODELS.has(input.declaration.model) && Array.from(text.text.trim()).length < 15) {
+  if (Array.from(text.text.trim()).length < 15) {
     throw new GenerationValidationError(`${input.declaration.model} requires input of at least 15 Unicode code points`);
   }
 }

src/builtins.ts

@@ -708,20 +708,14 @@ function geminiImageModel(
   };
 }
 
-function qwenTtsModel(
-  model: string,
-  title: string,
-  description: string,
-  options: { minimumTextCodePoints?: number } = {},
-): GenerationModelDeclaration {
-  const text = options.minimumTextCodePoints
-    ? "这是一段长度足够并且表达清晰自然的语音合成测试文本。"
-    : "这是一次清晰自然的语音合成测试。";
+const QWEN_TTS_TEXT = "这是一段长度足够并且表达清晰自然的语音合成测试文本。";
+
+function qwenAudio31TtsModel(model: string, title: string, tier: string): GenerationModelDeclaration {
   return {
     schema: MODEL_SCHEMA,
     model,
     title,
-    description,
+    description: `Modes: voice_prompt design OR one-reference clone. Text: >=15 Unicode code points. Tier: ${tier}. Conflict: ask user; never combine/reinterpret. Dependency: clone prior generated audio.`,
     adapter: { type: "openai.audioSpeech" },
     content: {
       input: [
@@ -730,9 +724,7 @@ function qwenTtsModel(
           required: true,
           min: 1,
           max: 1,
-          description: options.minimumTextCodePoints
-            ? `Exactly one non-empty text block to speak, with at least ${options.minimumTextCodePoints} Unicode code points.`
-            : "Exactly one non-empty text block to speak.",
+          description: "Exactly one non-empty text block to speak, with at least 15 Unicode code points.",
         },
         {
           type: "audio",
@@ -757,7 +749,7 @@ function qwenTtsModel(
         title: "Voice design",
         request: {
           model,
-          content: [{ type: "text", text }],
+          content: [{ type: "text", text: QWEN_TTS_TEXT }],
           meta: { voice_prompt: "一位沉稳干练的男性播音员声音，吐字清晰有力" },
         },
       },
@@ -766,7 +758,7 @@ function qwenTtsModel(
         request: {
           model,
           content: [
-            { type: "text", text },
+            { type: "text", text: QWEN_TTS_TEXT },
             { type: "audio", source: { type: "url", url: "https://example.com/reference.mp3" } },
           ],
         },
@@ -776,22 +768,11 @@ function qwenTtsModel(
 }
 
 const audioSpeechModels = [
-  qwenTtsModel(
-    "qwen-tts",
-    "Qwen TTS",
-    "Modes: voice_prompt design OR one-reference clone. Default: unspecified Qwen design. Text: any length. Conflict: ask user; never combine/reinterpret. Dependency: clone prior generated audio.",
-  ),
-  qwenTtsModel(
-    "qwen-audio-3.0-tts-plus",
-    "Qwen Audio 3.0 TTS Plus",
-    "Modes: voice_prompt design OR one-reference clone. Text: >=15 Unicode code points. Conflict: ask user; never combine/reinterpret. Dependency: clone prior generated audio.",
-    { minimumTextCodePoints: 15 },
-  ),
-  qwenTtsModel(
-    "qwen-audio-3.0-tts-flash",
-    "Qwen Audio 3.0 TTS Flash",
-    "Modes: voice_prompt design OR one-reference clone. Text: >=15 Unicode code points. Conflict: ask user; never combine/reinterpret. Dependency: clone prior generated audio.",
-    { minimumTextCodePoints: 15 },
+  qwenAudio31TtsModel("qwen-audio-3.1-tts-flash", "Qwen Audio 3.1 TTS Flash", "default; Alibaba's low-latency tier"),
+  qwenAudio31TtsModel(
+    "qwen-audio-3.1-tts-plus",
+    "Qwen Audio 3.1 TTS Plus",
+    "Alibaba's quality tier (audiobooks, dubbing); only when user or external policy selects it",
   ),
 ] satisfies GenerationModelDeclaration[];