Multimodal APIs
Text, vision, image, embeddings, audio, and speech—with honest endpoints
KeepRouter presents multiple model modalities in one catalog while preserving the endpoint and price unit each operation requires. Unified discovery does not flatten unlike payloads into a misleading universal request.
Last reviewed 2026-08-15 · Editorial review: KeepRouter Editorial
One catalog, several operation families
The public catalog labels each model's modality and links to the endpoint used to call it. Chat and vision-capable conversational models may use Chat Completions, Responses, or Messages when configured. Embedding models accept vectorization requests. Image models use image-generation routes. Speech and transcription models use audio-specific operations.
Why route-specific examples are safer
Each modality has different required fields, binary or URL inputs, output shapes, size limits, and billing units. A generic chat example on an image model page would look simple and fail in production. KeepRouter model pages only show a generic request example when the modality has a safe, supported pattern; otherwise they point to the API reference.
Price units follow the work
Text-like models are commonly metered by input and output tokens, with cached-input detail when reported. Image generation may be priced per image request. Audio operations can have their own units and payload constraints. Compare like with like and read the price label, not only the number.
Build a modality-aware client layer
Keep model configuration together with route, input validator, output parser, timeout, maximum payload size, and expected price unit. Do not let a UI selector swap a chat model for an image model while reusing the same request body. A typed operation registry prevents that class of failure.
Validate content handling
Images, audio, and documents create data-handling questions beyond text tokens. Check the selected model maker's policy, any routing intermediary involved, payload retention, regional constraints, and your own consent requirements. KeepRouter's security page describes gateway logging and transmission boundaries; it does not replace the upstream policy.
Start from the model page
Filter the catalog by modality, open a model detail page, confirm the endpoint and current price unit, then use the corresponding API docs. Keep the model ID and operation type together in configuration so later catalog changes do not turn into runtime guesswork.
Frequently asked questions
Do all modalities use /v1/chat/completions?
No. Specialized modalities use route-specific endpoints. The model page identifies the supported operation.
Are all models billed per token?
No. The price unit follows the modality; for example, some image models are priced per generated image.
Can one key call different modalities?
A key can call models allowed by its policy, but each request must use the correct route and payload for that modality.
Where should I verify payload limits?
Use the KeepRouter API reference, the model page, and the selected model maker's official documentation for provider-specific limits.