A non-text content block (image, tool_result, document, etc.).
By default, built-in estimators ignore non-text blocks — token
estimates for multimodal requests should then be treated as a lower
bound. To charge a fixed reservation per opaque block instead, use
createModelAwareTokenEstimator's opaqueBlockTokens option, or
carry a caller-computed total in LLMRequest.extraInputTokens, or
provide a custom estimator.
A non-text content block (image, tool_result, document, etc.).
By default, built-in estimators ignore non-text blocks — token estimates for multimodal requests should then be treated as a lower bound. To charge a fixed reservation per opaque block instead, use
createModelAwareTokenEstimator'sopaqueBlockTokensoption, or carry a caller-computed total inLLMRequest.extraInputTokens, or provide a custom estimator.