Inference operations and billing units
Choose the correct ShareAI endpoint, capabilities and pricing units for each kind of model.
On this page
Choose an operation, then a model#
ShareAI supports text chat and Responses, embeddings, images, asynchronous video, speech, structured decisions, rerank, transcription and audio generation. Input/output modalities describe the media a model understands or produces; the operation selects the endpoint and billing contract. A model can support multiple operations. Availability requires a configured, healthy executor and an active tariff for that exact operation.
- Generate images
- Generate videos
- Generate speech
- Transcribe audio
- Make structured decisions
- Rerank documents
- Generate audio
- Embeddings
Read the units before running#
Model pages publish unit prices, variants and volume or graduated bands. Tokens, characters, images, megapixels, seconds, documents and requests remain distinct. Embeddings price input only: vectors and vector dimensions have no output price. Media is never counted as text merely because it is transported as base64.
Creator and Provider responsibilities#
Creators define model operations, input/output capabilities and sharing or licensing terms. Providers configure their own operation tariffs for eligible models they serve. Creator licensing terms and inference prices are separate settings. A fallback route may call an external vendor; the ShareAI API, billing and usage contracts remain provider independent.
Authentication, privacy and retries#
All generation routes require a Bearer API key or approved OAuth access token. Use the same originating credential for a video job and its content. Keep credentials in server-side environment variables. ShareAI implements the documented subset: unknown fields are rejected before admission. Do not assume every upstream SDK option is accepted. Only video can explicitly opt into temporary result retention; other operations retain their privacy policy.
Response formats#
Chat and Responses may stream where documented. Embeddings, images, decisions, rerank, transcription and audio generation return bounded JSON. Speech returns binary audio. Video returns HTTP 202 and must be polled; status and content reads do not trigger another generation or ingestion fee.
Text APIs#
Last updated November 3, 2026