Changelog

Platform and API changes that affect how you build on Atlas Cloud.

2026-08

• API keys are revealed once — a new key is shown in full exactly once at creation and masked afterwards. Existing keys can be rotated from the API Keys page.
• Per-model protocol support is published — every LLM declares which protocols it accepts, so you can see whether a model expects Chat Completions, Responses, Anthropic Messages, or Gemini generateContent. See LLM API Protocols.
• Audio input for multimodal LLMs — models that declare audio support accept audio parts alongside text, images, and video.
• Auto top-up improvements — a prompt before your balance runs out, a balance panel in the navigation bar, and payment actions restricted from ordinary team members.
• Error details in history — usage history now shows the returned error message and raw response body, so structured results such as transcripts and lyrics are fully readable.

2026-07

• Webhook callbacks for audio jobs — audio generation can notify your endpoint on completion instead of being polled. See Webhooks.
• Spend breakdown by API key and member — analytics splits spend by key, member, and model over a chosen date range, with CSV export and explicit UTC boundaries.
• Multiple teams — create up to 5 teams, transfer cash balance from a personal account into a team, and assign member aliases. See Team Accounts.
• Time-of-use pricing — model pages show peak and off-peak rates where a model is priced that way.
• Per-model sampling parameters — the API reference renders only the parameters a given model actually accepts.
• Cash and bonus balance shown separately — the balance card distinguishes paid credit from granted credit.

2026-06

• Speech synthesis support — the platform gained a text-to-speech capability on POST /api/v1/model/generateAudio, billed per character. See Audio Models.
• Transcription support — speech-to-text runs through the same asynchronous pipeline and returns text rather than a file URL.
• 3D generation support — a new output category with GLB, OBJ, and USD-family formats, served through the image endpoint.
• Self-service invoices — request or re-issue an invoice for a transaction and download the PDF from billing history.
• OSS Program — open-source maintainers can apply for sponsored credits at atlascloud.ai/oss-program.
• Modality flags come from the API — input_modalities and output_modalities describe what each model accepts and returns.
• Media inputs by URL — reference images, video, and audio can be supplied by link instead of uploading a file.

2026-05

• Console renamed to Dashboard — the product surface was renamed throughout.
• CLI, MCP Server, and Skills — three developer integrations shipped together. See CLI, MCP Server, and Skills.
• Coding Plan subscriptions — subscribe in-product, toggle auto-renew, and create a dedicated Coding Plan key. See Coding Plan.
• Team roles and balance transfer — invite members, assign roles, and move balance between a personal account and a team.
• Batch generation — queue several runs of the same model at once.

2026-04

• Workflow Studio — build and run multi-step generation workflows on a canvas, with a shared template library and parallel execution.
• API key scopes — keys are created with a scope, and the key list shows which scope each one carries. See API Keys.
• Credit overdraft — accounts with an approved credit limit can continue past a zero balance instead of being blocked.
• Conditional schema validation — dependent and mutually exclusive parameters are validated before a request is submitted.

2026-03

• Image and video moved to asynchronous billing — media jobs are charged when the prediction settles rather than at submission, so failed jobs are not billed. See Model Billing.
• Rebuilt API reference — per-language code samples with parameter comments and a copy-as-Markdown action.
• Nested request schemas supported end to end — complex nested parameters are handled in the form, the parameter table, and validation.
• Billing address and invoice history — save a billing address and browse past invoices.

2026-02

• x-api-key accepted — the Anthropic SDK works against Atlas Cloud without a custom auth header, in addition to Authorization: Bearer. See LLM API Protocols.
• Discounts visible in the API — pricing responses carry both the list price and the effective price.
• Cached-token and tiered pricing labelled — model pages show cached-token pricing and context-length tiers where they apply.
• 402 on insufficient balance — generation requests return 402 when funds run out. See Errors & Rate Limits.
• Add an email to a social login — accounts created through Google or GitHub can bind an email address, which is required for billing. See Sign Up & Sign In.

2026-01

• Console moved into the main site — Dashboard, Billing, Usage History, and API Keys now live under atlascloud.ai/console with a single session across the site. There is no separate console domain to sign in to.
• Auto top-up and saved payment methods — store cards, order them, and set an automatic recharge threshold.
• Google One Tap sign-in — sign in or sign up without leaving the page.
• Dark mode — a site-wide theme toggle.
• Polling retries on network errors — a dropped connection while polling no longer fails the run.

2025-12

• Multiple images per response — image models that return more than one result display and download all of them.
• Reference-driven video — reference inputs became a first-class parameter type in the catalog and forms.
• Compare mode — run the same conversation against several models side by side.
• Refunds on failed generations — the error state indicates when a failed job has been refunded.
• 20-language localization — the site, model pages, and dashboard were localized.
• Data deletion and account closure policy published. See Data Deletion.

2025-11

• Request history on model pages — image and video model pages keep a history of your runs with prompt, status, and download.
• Schema-validated JSON editor — edit the raw request body with schema validation, inline errors, and live price calculation.
• Responses API support — selected models can be called through POST /v1/responses.
• Site-wide search across models, families, and collections.
• Upload limits surfaced — upload fields show progress and enforce a size limit before the request is sent.

2025-10

• API tab on model pages — a parameter reference table, syntax-highlighted request samples, and OpenAI SDK examples on every model page.
• Audio and end-frame inputs — generation forms accept audio files and an end frame for models that support them.
• Payment methods expanded — cards, WeChat Pay, Alipay, and cryptocurrency.
• Pricing pages — per-model pricing with live price calculation.
• Linkable filtered views — model list filters sync to query parameters.

2025-09

• Image and video model catalog — a dedicated browse surface with provider filters and per-model detail pages.
• Sessions moved to HttpOnly cookies — auth tokens are no longer readable from browser storage.
• Balance checked before submission — insufficient funds are detected before a generation request is sent, rather than surfacing as a failed job.

---

Looking for a change that is not listed? Contact support or check the Model Library for the current catalog.