Invalid or missing transcription model. Please configure a v

Error message

Invalid or missing transcription model. Please configure a valid model in the provider settings.

What it means

The openai-compatible transcription page validates the resolved model before calling hearingStore.transcription: modelToUse comes from providerConfig.model, falling back to the reactive model.value, and must pass isValidTranscriptionModel(). That validator accepts known transcription models (whisper-*, gpt-4o-*-transcribe) and custom names, but rejects obvious chat models (names containing 'gpt-4' without 'transcribe'/'whisper'). The error means the resolved model was empty or classified as a chat model.

Source

Thrown at packages/stage-pages/src/pages/settings/providers/transcription/openai-compatible-audio-transcription.vue:89

// Check if API key is configured
const apiKeyConfigured = computed(() => !!providers.value[providerId]?.apiKey)

// Generate transcription
async function handleGenerateTranscription(file: File) {
  const provider = await providersStore.getProviderInstance<TranscriptionProviderWithExtraOptions<string, any>>(providerId)
  if (!provider)
    throw new Error('Failed to initialize transcription provider')

  // Get provider configuration
  const providerConfig = providerStore.getProviderConfig(providerId)

  // Get model from configuration or use the reactive model value
  const modelToUse = providerConfig.model as string | undefined || model.value

  // Validate model - throw error if no valid model configured
  if (!modelToUse || !isValidTranscriptionModel(modelToUse)) {
    throw new Error(`Invalid or missing transcription model. Please configure a valid model in the provider settings.`)
  }

  return await hearingStore.transcription(
    providerId,
    provider,
    modelToUse,
    file,
    'json',
  )
}

// Use the composable to get validation logic and state
const {
  t,
  router,
  providerMetadata,
  isValidating,
  isValid,

View on GitHub (pinned to 677329427f)

Solutions

Open provider settings and pick a valid transcription model (whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe, or your endpoint's whisper-compatible name), then save.
If your endpoint uses a custom transcription model name, ensure it does not contain 'gpt-4' without also containing 'transcribe' or 'whisper'.
Persist an explicit default model in the provider's default config so modelToUse is never empty.
Check onMounted's auto-reset (line 172) ran — an invalid persisted model should be reset to 'whisper-1'; if it did not, clear the stored model value and re-select.

Example fix

// before
const modelToUse = providerConfig.model as string | undefined || model.value
if (!modelToUse || !isValidTranscriptionModel(modelToUse)) {
  throw new Error(`Invalid or missing transcription model. Please configure a valid model in the provider settings.`)
}

// after — fall back to whisper-1 and warn instead of hard-failing
const modelToUse = providerConfig.model as string | undefined || model.value || 'whisper-1'
if (!isValidTranscriptionModel(modelToUse)) {
  throw new Error(`Invalid transcription model "${modelToUse}". Choose a transcription model (e.g. whisper-1) in provider settings.`)
}

Defensive patterns

Strategy: validation

Validate before calling

const modelToUse = providerConfig.model as string | undefined || model.value
if (!modelToUse || !isValidTranscriptionModel(modelToUse)) {
  // block generation and prompt model selection; do not call hearingStore.transcription
}

Type guard

function isValidTranscriptionModel(modelName: string | undefined | null): boolean {
  if (!modelName)
    return false
  if (VALID_TRANSCRIPTION_MODELS.includes(modelName))
    return true
  if (modelName.includes('gpt-4') && !modelName.includes('transcribe') && !modelName.includes('whisper'))
    return false
  return true
}

Try / catch

try {
  await handleGenerateTranscription(file)
}
catch (error) {
  if (errorMessageFrom(error).includes('Invalid or missing transcription model')) {
    // focus the model selector in provider settings
  }
  throw error
}

Prevention

Validate the model in the settings form on save, not only at generation time.
Default the openai-compatible provider to 'whisper-1' so modelToUse is never empty.
Re-run the onMounted auto-reset check after programmatic config changes.

When it happens

Trigger: providerConfig.model and model.value are both empty (no model ever selected/saved), or the saved model is something like 'gpt-4o' / 'gpt-4-turbo' which fails the chat-model rejection branch at openai-compatible-audio-transcription.vue:155.

Common situations: User previously saved a chat model on a shared openai-compatible provider config; a fresh install where no default model is set for the openai-compatible provider (unlike the official page it has no hardcoded fallback here); custom endpoint exposing only chat models; stale persisted model value from an older config layout.

Related errors

AI-assisted analysis of moeru-ai/airi@677329427f (2026-08-18). Data as JSON: /api/errors/37138966813cda71. Report an issue: GitHub.