Changelog

Submit and follow up

• Open a ticket from the console and follow the conversation there; five categories narrow down to the exact problem

• Rich text and inline images in the body — screenshots can be pasted straight in

• Attach ledger entries, invoices, top-up orders or call records so support does not have to ask for IDs

Where you can start one

• Every billing ledger row can raise a billing question with that record attached

• Call log details can raise feedback about that specific call

• Workspace members page can contact support about that workspace

Refunds

• Refund tickets can reference the top-up order; the payout and processing cost are both shown

Video generation

• New video modality: submit a task, poll for the result, artifacts are stored by the platform

• Alibaba Wan3.0, HappyHorse, MiniMax H3 and H3-Max, Doubao Seedance 2.0 / 2.0-fast / 2.0-mini / 2.5

• Video is billed by output duration; Seedance settles on the token count the upstream returns

Image and text

• Qwen-Image 3.0 for image generation, plus Qwen text models

• xAI Grok: text, in-chat image generation and image-to-image

Billing and usage

• Cache hits are split into explicit and implicit tiers, each priceable separately

• Per-call server tool billing expanded from one item to a full table

• Public pricing shows every tier, not just the entry price

Console

• Log details play generated videos inline; the list gains a video output column

• Filters are written into the URL — refresh, share or hit back and they survive

Off-peak pricing

• Model prices can be configured per time window — the same model can cost differently at different hours

• Comparison and pay-as-you-go tables show every window side by side, in your own timezone

• Billing details and call logs state which window a charge was priced under

Top-ups

• Top-ups now go through Stripe Checkout; Alipay is available again

• The amount snaps back into the allowed range when the field loses focus

Billing

• Ledger and API key tables are paginated

• Truncated upstream responses are no longer recorded as successes

Playground

• Chat, generate images and generate video in the console — credentials are managed by the platform, no key setup needed

• Compare several models side by side in draggable columns

• Attach images, audio, video or PDFs according to what each model accepts

• Image and video have their own composers: a draggable aspect-ratio panel, a duration slider and material slots; a running job survives a page refresh and can be cancelled

• The conversation and the selected model live in the URL, so a shared link carries the same setup

Model comparison

• A new page compares several model-and-provider pairs on price, performance, endpoints and features

Console

• Rows per page are remembered per table; several lists moved to server-side pagination

• Model pickers show model icons, and long model names no longer break the card

New models

• Gemini can generate images, reachable through the OpenAI Images request format

• DeepSeek is available; its cache usage is counted under its own field names

Behaviour

• Upstream payment failures now have their own error type: no retry, and no misleading prompt to top up your own balance

• Web search counts are now collected on the Anthropic and Responses paths too

Billing

• Web search is billed per call

• Removed a batch of price components that were configurable but never actually charged

• The pay-as-you-go table's columns follow the components that really bill

Docs

• Every parameter in the API reference now has a property description, nested ones included

• Model pages no longer restate prices — pricing lives in one place

• Supported aspect ratios widened to 14

Fixes

• Media usage could be dropped entirely, showing a zero bill; fixed

• Image tokens are no longer charged a second time at text rates

APIs

• POST /v1/images/generations and POST /v1/images/edits are live; edits accepts multipart

• POST /v1/tasks submits long-running jobs and returns results by id, with Prefer: wait, Idempotency-Key and cancellation

• Synchronous streaming image generation

Webhooks

• Completed tasks can call back to your endpoint, signed

• The signing secret is available in the console, and the docs include a verification example

Console

• Logs cover images: a dedicated output column, and details split into a rendered view and raw JSON

• Amounts are shown at full precision in both list and detail

Fixes

• A reservation left behind by a crash no longer freezes your balance forever

• Failed image requests no longer appear twice in the log

Image APIs

• Supports both OpenAI Images and Gemini request formats

• Image and text requests use the same API keys and routing settings

Task processing

• Long-running jobs support queueing and polling

• Busy upstreams return Retry-After so clients can schedule retries

Billing

• Pricing is based on image count and dimensions

• An estimate is reserved on submission; final usage is settled on completion and the remainder is released

• Reasoning text produced before the image is included in usage

Payments

• Supports manual and automatic top-ups

• Automatic top-ups can be triggered when balance falls below a configured threshold

• Refunds are returned to the original credit lot

Billing

• The console separates token usage from monetary charges

• The cost page provides transaction details and the source of each charge

Public pricing

• New overview, pay-as-you-go, subscription and promotion pages

• The pricing table can be viewed by model, vendor and context tier, with sorting and pagination

Parameter support

• Lists supported parameters and accepted values for each provider

• Model pages show provider-specific parameters available for that model

• temperature limits follow each platform's actual range

Request feedback

• Parameters that are not forwarded are listed in the response

• Parameters that can change pricing are rejected with a specific reason until a pricing rule is configured

Pricing

• Price tables can define tiers based on request parameters

MiniMax

• Added a first-party MiniMax route alongside other MiniMax providers

• Routing preferences can be limited to first-party providers

• Supports thinking mode and cached-token reporting

• Cached tokens are billed at MiniMax's official rates

Promotion rules

• Promotion windows combine time of day, days of the week and days of the month

• A promotion can define multiple discount tiers

Price display

• Model lists and detail pages show both the discounted price and the original price

Performance data

• Added throughput, latency, availability and usage charts

• Time ranges cover the last 1 day, 7 days and 30 days

• The availability bar splits the selected range into 48 cells, with exact values on hover

• Select a provider to view its performance separately

Quick start

• Model cards and detail pages now include an API request entry point

• Choose a key, protocol and tool in three steps to generate a ready-to-use configuration

Time display

• Charts and records use the visitor's local time zone

API reference

• Grouped by OpenAI Chat, OpenAI Responses, Anthropic and Gemini protocols

• Each group includes list-models endpoints, parameter tables and request examples

• Added documentation for the @ routing-strategy suffix

Integration guides

• Added installation, configuration and troubleshooting steps for CC Switch, Sider, Pi and other tools

• API keys can be viewed in the console at any time instead of only once at creation

Policies

• Added a refund policy page

Conversion coverage

• All 12 paths between OpenAI Chat Completions, OpenAI Responses, Anthropic Messages and Gemini are available

• An existing SDK can call models served through another protocol

Parameters and errors

• Supports parallel_tool_calls and caller identity

• Unsupported content blocks return a 400 with details about what could not be converted

• Compatibility docs list field mappings for each conversion path

Routing preferences

• Provider scope can be limited to first-party routes or include all routes

• Ordering strategies include price, throughput, latency and availability

• Append a suffix such as @price to the model name to select a strategy for one request

• Requests in the same session prefer the same route to preserve context-cache hits

Usage limits

• Members and API keys can have daily, weekly, monthly and lifetime limits

• New requests are blocked before reaching an upstream once a limit is reached

Configuration scope

• Personal and organization accounts keep separate routing settings

Sign-in

• The sign-in page now includes a live view of the model catalog

API access

• The gateway provides multiple domains; any of them can be used as the API base URL

Status guidance

• The API and console show a specific reason when an account or workspace is restricted

• Messages include the available recovery action

• Restrictions apply to new requests; existing API keys and historical data remain available

Notifications

• The message center separates model launches, platform updates and important notices

• Supports marking individual items or all items as read

Changelog

• Added a complete history at /changelog

• Provides an RSS feed

• Backfilled entries do not create duplicate unread alerts

Uploads

• Avatars and content images upload directly to object storage

Loading

• Pages load an image size appropriate for the display area to reduce transfer size

Organizations and workspaces

• An organization can contain multiple workspaces

• The top switcher moves between personal and organization workspaces

• Balance and billing belong separately to personal or organization accounts

Members

• Invite members by email and assign roles

• API keys belong to the member who created them; administrators can view all keys in the organization

Usage

• Organization usage can be viewed by member and workspace

Credits

• Supports account balance, trial credits and expiring credit lots

• Credits with the earliest expiry are consumed first

Request billing

• Available balance is checked before a request is sent

• Insufficient balance returns a 402 with a top-up URL

• Charges are based on completed usage; upstream failures are not billed

Request logs

• Records model, protocol, tokens, total duration, time to first token and cost breakdown

• Failed requests identify whether the issue came from the request, platform or upstream

• Filter by time, model, API key, status, request ID and other fields

Usage statistics

• Shows total requests, tokens, cost and latency

• Usage and cost trends can be grouped by model and API key

Search coverage

• Press ⌘K to open site search

• Search guides, API reference and the model catalog together

Search history

• The search panel keeps recent queries

Sections

• Documentation is divided into guides, API reference and integrations

API reference

• API pages are generated from the OpenAPI specification

• Each endpoint includes request examples in eight languages

Raw content

• Documentation pages expose a raw-content URL

• llms.txt is available for models and developer tools

Model quotes

• /v1/models returns provider quotes with units and currency

• Model lists show the lowest current price and its conditions

Tiered pricing

• Supports different rates based on context length

• Line items are not rounded independently; settlement is rounded once at the total

Interrupted requests

• Disconnected requests are billed for content already returned

Failover

• A failed provider request automatically moves to the next available route

• Request logs record the full attempt chain and each failure reason

Circuit breaking

• Routes with repeated failures stop receiving traffic automatically

• Periodic probes return recovered routes to service

Gemini

• Added the native /v1beta/models/{model}:generateContent endpoint

• Gemini can serve requests received through other protocols

OpenAI Responses

• Available models on /v1/responses increased from 8 to 20

Model information

• Added model catalog and detail pages

• Shows vendor, context length, pricing and supported protocol endpoints

Model aliases

• A model can be called by multiple accepted names

• Legacy model IDs redirect to the canonical page

Pricing

• Shows price tiers based on context length

Sign-in

• Supports email verification codes

• Supports Google and GitHub sign-in

API keys

• Supports API key creation and verification

Interface

• Available in English and Chinese

• Supports light and dark themes