Caching and batch behavior 

Caching 

• Embeddings: 24h cache by default; EmbeddingOptions::noCache() opts out.
• Plain completion/chat: configuration answer_cache_ttl opts in; off by default and subject to privacy and path restrictions.
• Vision: separate from the plain-answer cache path.
• Translation: The translator path opts in with a positive TranslationOptions::cacheTtl and needs the cache service. LLM translation and polishing also need a known model/configuration context for their cache key; other translators do not require a default LLM configuration. See Translation options.

Batch processing 

Image arrays provide a convenient batch interface. Each image makes its own provider request, in order; batching does not reduce the number of requests.

// One provider request per image, with results in the same order
$altTexts = $visionService->generateAltText($imageUrls);