vercel/ai · error · UnsupportedFunctionalityError

AI_UnsupportedFunctionalityError

AI_UnsupportedFunctionalityError

Error message

Alibaba embedding outputType 'sparse' is not supported because AI SDK embeddings require dense number arrays. Use 'dense' or 'dense&sparse' instead.

What it means

Alibaba embedding models can return sparse (index/value) vectors, but the AI SDK embed API only accepts dense number[] arrays. The provider therefore proactively rejects outputType 'sparse' with an UnsupportedFunctionalityError instead of returning data the SDK cannot represent. Use 'dense' or 'dense&sparse' (the SDK surfaces only the dense portion).

Source

Thrown at packages/alibaba/src/alibaba-embedding-model.ts:89

  > {
    if (values.length > this.maxEmbeddingsPerCall) {
      throw new TooManyEmbeddingValuesForCallError({
        provider: this.provider,
        modelId: this.modelId,
        maxEmbeddingsPerCall: this.maxEmbeddingsPerCall,
        values,
      });
    }

    const alibabaOptions = await parseProviderOptions({
      provider: 'alibaba',
      providerOptions,
      schema: alibabaEmbeddingModelOptions,
    });

    // TODO: Explore first-class sparse embedding support in AI SDK core.
    if (alibabaOptions?.outputType === 'sparse') {
      throw new UnsupportedFunctionalityError({
        functionality: "Alibaba embedding outputType 'sparse'",
        message:
          "Alibaba embedding outputType 'sparse' is not supported because AI SDK embeddings require dense number arrays. Use 'dense' or 'dense&sparse' instead.",
      });
    }

    const {
      responseHeaders,
      value: response,
      rawValue,
    } = await postJsonToApi({
      url: `${this.config.baseURL}/services/embeddings/text-embedding/text-embedding`,
      headers: combineHeaders(this.config.headers?.(), headers),
      body: {
        model: this.modelId,
        input: {
          texts: values,
        },

View on GitHub (pinned to 69428b1f8b)

Solutions

  1. Change the providerOptions outputType to 'dense' or 'dense&sparse'
  2. If sparse vectors are required, call the Alibaba/DashScope API directly outside the AI SDK
  3. Combine dense embeddings from the SDK with a separate sparse retrieval layer (e.g. BM25) instead

Example fix

// before
const { embedding } = await embed({
  model: alibaba.embedding('text-embedding-v4'),
  value: 'hello',
  providerOptions: { alibaba: { outputType: 'sparse' } },
});
// after
const { embedding } = await embed({
  model: alibaba.embedding('text-embedding-v4'),
  value: 'hello',
  providerOptions: { alibaba: { outputType: 'dense' } },
});
Defensive patterns

Strategy: validation

Validate before calling

if (opts.alibaba?.outputType === 'sparse') throw new Error('Use dense or dense&sparse outputType for AI SDK embeddings');

Type guard

function isDenseOutput(o: unknown): boolean {
  const t = (o as { alibaba?: { outputType?: string } })?.alibaba?.outputType;
  return t !== 'sparse';
}

Try / catch

try {
  await embed({ model, value, providerOptions });
} catch (e) {
  if (UnsupportedFunctionalityError.isInstance(e)) {
    // fall back to dense outputType
  }
}

Prevention

When it happens

Trigger: Calling embed()/embedMany() with an alibaba.embedding(modelId) model while passing providerOptions alibaba.embeddingModelOptions with outputType: 'sparse' in doEmbed.

Common situations: Developers porting code that used Alibaba's native sparse vectors for keyword-style retrieval, or copying DashScope docs that recommend sparse output for search ranking, then wiring it into the AI SDK.

Related errors

AI-assisted analysis of vercel/ai@69428b1f8b (2026-08-30). Data as JSON: /api/errors/6be812bf098226e2. Report an issue: GitHub.