Changelog

Model removals are also listed on Deprecations.

2 November 2026

Docs for AI agents. New llms.txt and llms-full.txt, a Markdown copy of every docs page (add .md to its URL), one Markdown file per model (/docs/models/<model id>.md) and one merged OpenAPI file. New guide: For AI agents.
New guide: AI-generated content. Outputs keep the labels the model adds (C2PA metadata, watermarks), byte for byte.
Jobs: webhooks, uploads and clearer fields. Pass webhook_url on POST /v1/jobs to be called when a job finishes (Standard Webhooks signatures). Upload input files with POST /v1/files. Jobs now return id, status, outputs and error; the older fields still work. Jobs.
Jobs. GET /v1/jobs/{id} now returns only the fields you need. A finished job shows generating until its files are copied to our storage (up to 10 minutes), so result links always point to lasting copies. Error and fail messages no longer contain third-party links.
Docs pages have canonical links, and the site has a robots.txt and sitemap.xml.

Text holds. A text request now reserves its longest possible answer before it runs: max_tokens if set, otherwise 8192 tokens, at the output price. You still pay only the real tokens; the rest is released when the reply ends.
Model removed: google/gemini-2-5-pro-tts. See Deprecations.
Console cleaned up: clearer type, colour and wording throughout.

Legal pages published under /legal: terms, privacy, acceptable use, refunds, cookies, model provider terms.
Console redesign. The app sidebar slides out on narrow screens.

Docs site with a searchable Models index and a page per model (price per setting, every input field, an example), the API reference with a playground, and Apps guides for Claude Code, Codex CLI, Cherry Studio, CC Switch and DeepChat.

Media models. Image, video, music and speech models through one Jobs endpoint, POST /v1/jobs and GET /v1/jobs/{taskId}, and through OpenAI-style /v1/images/generations and /v1/videos. Each job reserves its expected cost plus 20% and settles at the real cost; failed jobs cost nothing.
Text models on /v1/chat/completions, /v1/responses and /v1/messages.
MCP server at /mcp, with sign-in for Claude.ai and ChatGPT connectors, or your key.
Studio in the console: Image, Edit image, Video, Music, Speech, Motion and avatars, Tools (upscale, remove background, isolate voice) and a Gallery, with follow-up actions on finished results.
Rate limits: 120 requests a minute per account, 60 of them successful. See Rate limits.
Retention: uploaded files are kept 1 day, generated files 7 days. See Data and logging.
Content rules: safety filters always on; sexual content involving minors is refused.
Sign-in with a 6-digit email code; sign-up needs a verified email.