Multimodal
Beyond chat: image, audio & video
Chat completions handle text and, via image understanding, image input. Everything else — generating an image, turning text into speech, transcribing audio, generating video — is a dedicated OpenAI-compatible endpoint, each billed with the same 1% fee, trace row, and credit-wallet debit as every chat request. A few input modalities (PDFs, audio, video) aren't handled inside chat yet — those pages below say so plainly rather than pretending they work.
The eight pieces
Image understanding
Supportedimage_url content parts in chat
Image generation
SupportedPOST /v1/images
PDFs
Not yettype:file content parts in chat
Audio
Not yetinput_audio content parts in chat
Video understanding
Not yetvideo content parts in chat
Video generation
SupportedPOST /v1/videos, async job/poll
Text-to-speech
SupportedPOST /v1/audio/speech
Speech-to-text
SupportedPOST /v1/audio/transcriptions
Which models, and pricing
The four output endpoints (image, video, TTS, STT generation) serve a small, hand-curated set of models spanning multiple providers (OpenAI, Google, MiniMax, ElevenLabs, Qwen) — not the full chat registry's ~300+ multi-provider failover catalog; each model is served direct by whichever provider it belongs to, one model per provider rather than automatic fallback across providers. See the "Image, audio & video" section on /models for the current list, provider, and per-unit pricing (per image, per character, per minute, per second).