vercel/ai · error · NoObjectGeneratedError

No object generated: could not parse the response.

Error message

No object generated: could not parse the response.

What it means

This NoObjectGeneratedError is thrown when the model's text output cannot be parsed as JSON at all (`safeParseJSON` fails). generateObject/streamObject expect the model to emit valid JSON matching the structured-output prompt/schema; prose, markdown fences, truncated output, or refusals all cause this error. The original parse error and raw text are attached as `cause` and `text`.

Source

Thrown at packages/ai/src/generate-object/parse-and-validate-object-result.ts:33

 * @param result - The result string to parse and validate
 * @param outputStrategy - The output strategy containing validation logic
 * @param context - Additional context for error reporting
 * @returns The validated result
 * @throws NoObjectGeneratedError if parsing or validation fails
 */
async function parseAndValidateObjectResult<RESULT>(
  result: string,
  outputStrategy: OutputStrategy<any, RESULT, any>,
  context: {
    response: Omit<LanguageModelResponseMetadata, 'messages'>;
    usage: LanguageModelUsage;
    finishReason: FinishReason;
  },
): Promise<RESULT> {
  const parseResult = await safeParseJSON({ text: result });

  if (!parseResult.success) {
    throw new NoObjectGeneratedError({
      message: 'No object generated: could not parse the response.',
      cause: parseResult.error,
      text: result,
      response: context.response,
      usage: context.usage,
      finishReason: context.finishReason,
    });
  }

  const validationResult = await outputStrategy.validateFinalResult(
    parseResult.value,
    {
      text: result,
      response: context.response,
      usage: context.usage,
    },
  );

View on GitHub (pinned to 69428b1f8b)

Solutions

Inspect `error.text` and `error.cause` to see the raw output and parse failure; adjust the prompt to demand raw JSON only.
Use a model/provider that supports native structured output or JSON mode (e.g. provider `structuredOutputs`/`response_format` options).
Increase `maxOutputTokens` if the JSON is truncated (check `finishReason` for 'length').
Lower temperature and add 'respond with JSON only' instructions; use `experimental_repairText` (parseAndValidateObjectResultWithRepair path) to auto-fix near-miss JSON.
If using array/enum mode, confirm the schema wrapping is being sent (don't override the mode's JSON schema).

Example fix

// before
generateObject({ model: openai('gpt-4o'), schema, prompt: 'List 3 users', maxOutputTokens: 50 }); // truncated JSON

// after
generateObject({
  model: openai('gpt-4o', { structuredOutputs: true }),
  schema,
  prompt: 'List 3 users. Respond with JSON only.',
  maxOutputTokens: 1000,
  experimental_repairText: FixableJSONRepair,
});

Defensive patterns

Strategy: try-catch

Validate before calling

// after the fact, inspect attached diagnostics:
// error.text (raw model output), error.cause (parse error), error.finishReason ('length' => truncated)

Type guard

import { NoObjectGeneratedError } from 'ai';
function isNoObjectGenerated(e: unknown): e is NoObjectGeneratedError {
  return NoObjectGeneratedError.isInstance(e);
}

Try / catch

try {
  const { object } = await generateObject({ model, schema, prompt });
  return object;
} catch (e) {
  if (NoObjectGeneratedError.isInstance(e)) {
    console.error('raw text:', e.text, 'cause:', e.cause, 'finishReason:', e.finishReason);
    return retryWithJsonOnlyPrompt(); // or repair path
  }
  throw e;
}

Prevention

Use providers/models with native structured output or JSON mode.
Add 'respond with JSON only, no markdown' to prompts.
Set maxOutputTokens high enough for the full JSON.
Use experimental_repairText for near-miss JSON.
Check error.finishReason === 'length' to detect truncation.

When it happens

Trigger: Model returns non-JSON text (refusal, explanation, markdown code fences); output truncated by `maxOutputTokens`/`maxTokens`; in array/enum mode the model omits the required `elements`/`result` wrapper so the overall payload is invalid JSON; `output: 'no-schema'` where the prompt alone fails to elicit JSON.

Common situations: Models without native JSON/structured-output support; small token limits cutting off mid-JSON; temperature too high producing chatty preambles; using `generateObject` with `output: 'no-schema'` and a prompt that yields prose.

Related errors

AI-assisted analysis of vercel/ai@69428b1f8b (2026-08-30). Data as JSON: /api/errors/2623891521ced003. Report an issue: GitHub.