What shipped, newest first.
The docs pages now state what each endpoint accepts and link to the reference: the Support and limits notes, the Codex Desktop and Claude Desktop connection guides, and the Endpoint index were rewritten to the point.
The Video and Image generation pages open with a table per model: durations, resolutions, ratios, first and last frame, reference images, video and audio, audio output, options and price basis. Each feature then has its own section with a request you can copy. A new Thinking and max_tokens page lists which models think, the switch and effort fields for each, caps that leave room for the answer, how the hold and billing work, and how reasoning streams. Kling Motion Control takes a base64 character image as well as a URL.
Seedream 4.0, 4.5, 5.0 Lite and 5.0 Pro take up to 10 reference images in input_references. MiniMax H3 takes up to 9 reference images, one reference video and one reference audio. MiniMax H3 Max does not take references. Prices are on each model's page.
Each live API's page runs a real sample call with one of your keys. It uploads a short sample file, waits for the job to finish, and shows the result with what the call cost.
Every Tool opens on one page with About, Run and History. A run continues when you leave the page, and History lists your workspace's past runs until their files expire.
Usage split by source, workspace budgets, referrals, refund requests and payment methods.
Add an authenticator app as a second factor on your account.
Captions, Dubbing, Transcriber, Face Blur, Document Reader and more run on the same APIs you can call yourself, priced in credits per run.
Rotate a key with a 24-hour grace period for the old secret and set a rate limit per key.
Save Playground presets, ask for JSON output against a schema, attach images, and receive a webhook when a media job finishes.
Fixed-unit APIs such as OCR, transcription and translation, listed on the APIs page with their unit prices.