API Updates

Stay up to date with OmniMux gateway model launches and platform changes. The full callable catalog is on console pricing / Pricing.

New modelsImage

GPT Image Tiers

OmniMux now supports dedicated tier routing models gpt-image-2.5-flare and gpt-image-2.5-sunburst under the standard image generation endpoint.

Model specification

Endpoint: POST /v1/images/generations
Authentication: Bearer token (sk-...)
Tier models: gpt-image-2.5-flare (discounted tier), gpt-image-2.5-sunburst (economy tier)
Parameters: Fully compatible with baseline gpt-image-2.5

Request example

curl --request POST \
  --url https://api.omnimux.ai/v1/images/generations \
  --header 'Authorization: Bearer sk-...' \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "gpt-image-2.5-flare",
    "prompt": "A clean studio product photograph on white background",
    "n": 1
  }'

Parameter details are documented in the GPT Image Reference.

Video Task and Variant Models

OmniMux now supports dedicated per-call video task channels and H3 model variants under the standard video endpoint.

Model specification

Endpoint: POST /v1/video/generations
Polling endpoint: GET /v1/video/generations/{task_id}
Per-call models: seedance-2-0-task, seedance-2-5-task, minimax-h3-task (billed per task call)
H3 variants: minimax-h3-turbo (turbo rendering), minimax-h3-video (dedicated video pipeline)

Request example

{
  "model": "seedance-2-0-task",
  "prompt": "A cinematic drone shot of coastal mountains at sunset",
  "seconds": 5
}

Full parameter definitions are documented in the Seedance 2.0 Reference and MiniMax H3 Reference.

Links

New modelsAudio

Doubao Speech Recognition

The doubao-asr-bigmodel, seedasr-auc, and bigasr-auc models are now supported on OmniMux.

Model specification

Model IDs: doubao-asr-bigmodel, seedasr-auc, bigasr-auc
Protocol: OpenAI-compatible Audio Transcriptions (POST /v1/audio/transcriptions)
Authentication: Bearer token (sk-...)
Input format: Audio file multipart upload (wav, mp3, ogg, m4a)

Request example

curl --request POST \
  --url https://api.omnimux.ai/v1/audio/transcriptions \
  --header 'Authorization: Bearer sk-...' \
  --header 'Content-Type: multipart/form-data' \
  --form 'file=@audio.mp3' \
  --form 'model=doubao-asr-bigmodel'

Full parameter specifications are documented in the Doubao ASR Reference.

Seed Audio 1.0

The seed-audio-1.0 model is now supported on OmniMux under standard audio speech endpoints.

Model specification

Model ID: seed-audio-1.0
Protocol: OpenAI-compatible Audio Speech (POST /v1/audio/speech)
Authentication: Bearer token (sk-...)
Output limit: Up to 120 seconds per generation
Prompt limit: Up to 3000 characters

Request example

curl --request POST \
  --url https://api.omnimux.ai/v1/audio/speech \
  --header 'Authorization: Bearer sk-...' \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "seed-audio-1.0",
    "input": "Text to synthesize into speech",
    "voice": "zh_female_cancan",
    "response_format": "mp3"
  }' \
  --output output.mp3

Full parameter specifications are documented in the Seed Audio Reference.

Index TTS 2

The indextts-2 voice clone model is now supported on OmniMux.

Model specification

Model ID: indextts-2
Create endpoint: POST /v1/tasks/autodl
Polling endpoint: GET /v1/tasks/{task_id}
Artifact endpoint: GET /v1/tasks/{task_id}/artifacts
Authentication: Bearer token (sk-...)

Request example

{
  "model": "indextts-2",
  "prompt_text": "Hello, this is a cloned voice generation test.",
  "prompt_simple": "https://example.com/reference_audio.wav",
  "emo_control_method": "same_as_reference"
}

Full parameter specifications are documented in the Index TTS Reference.

Suno Sound Effects

The dedicated sound effect generation model suno-sounds is now supported on OmniMux.

Model specification

Model ID: suno-sounds
Endpoint: POST /v1/video/generations
Polling endpoint: GET /v1/video/generations/{task_id}
Authentication: Bearer token (sk-...)
Modality: Ambient and scene sound effects without vocal lyrics

Request example

curl --request POST \
  --url https://api.omnimux.ai/v1/video/generations \
  --header 'Authorization: Bearer sk-...' \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "suno-sounds",
    "prompt": "footsteps on gravel followed by heavy rain and distant thunder"
  }'

Details are available in the Suno API Reference.

Gemini 3.8 Flash

The gemini-3.8-flash model is now supported on OmniMux under the standard Gemini Chat Completions endpoint.

Model specification

Model ID: gemini-3.8-flash
Protocol: OpenAI-compatible Chat Completions
Endpoint: POST /v1/chat/completions
Authentication: Bearer token (sk-...)
Streaming: Supported via "stream": true

Request example

{
  "model": "gemini-3.8-flash",
  "messages": [
    {
      "role": "user",
      "content": "Hello"
    }
  ]
}

Full parameter schema and status codes are documented in the Gemini Complete API Reference.

What is new

Official Direct Channel: Added VolcEngine Ark direct route #73 for ByteDance Doubao seed-evolving.
Full Multimodal Reasoning: 1024K context window, 256K max output tokens, native deep thinking, tool calling, and structured outputs.
Transparent 1:1 Parity Pricing: Token billing expressions anchor official list prices ( $0.8824/M input,$ 4.4118/M output, $0.1765/M cache hit; FX 6.8).
Local OmniMux CLI Verification: Verified live invocation and pricing queries using the local omnimux CLI (omnimux models, omnimux pricing, omnimux tokens exec).

Documentation

Doubao API Reference: /en/api-reference/text-series/doubao/complete

What changed

Image-generation model pages now document the reference-image request field images (string[], image URL or base64) together with the accepted aliases image, image_urls and input_reference, which the gateway normalizes to images before the upstream call.
Each page states the limits its vendor publishes for that model: maximum count, minimum, accepted formats and per-file size ceiling.

Limits we do not state

Where a value is not published by the vendor, the page says so instead of carrying a number; nothing is rounded or copied from a sibling model.
Where no model spec exists yet, the reference-image note marks the capability unverified and states no limit at all.

Withdrawn figure

The previously stated per-file reference-image cap of 50 MB is withdrawn. In the vendor’s image guide that figure sits under mask requirements and covers the image being edited together with its mask; it is not a reference-image ceiling. gpt-image-2.5 therefore carries no per-file size limit.

Documentation

Image series model pages: /en/api-reference/image-series/models/*
GPT Image generation: /en/api-reference/image-series/gpt-image/generate

Links

BreakingImage

Required migration

Replace the legacy midjourney ID and every per-action ID (for example mj-v7-upscale, mj-v8-1-remove-bg) with one of two version IDs: mj-v7 or mj-v8-1.
Select the action in the request body: model_params.operation = generate | enhance | pan | outpaint | remove-bg | upscale | variation | edit | inpaint | remix | retexture | upload-paint. Omit it for plain generation.
Keep using POST /v1/images/generations; the request shape, the async task id and the polling behaviour are unchanged.

The old IDs are not compatibility aliases. An unknown or unpriced operation is rejected with 400, never silently downgraded to plain generation.

Pricing

Per-call prices are tiered by action, and the tiers are the same on both versions:

Action tier	Price	Credits
`generate`, `enhance`, `pan`, `outpaint`, `remove-bg`	USD 0.113446	1.134
`upscale`, `variation`, `edit`, `inpaint`, `remix`, `retexture`, `upload-paint`	USD 0.170169	1.702

Multipliers stack on the action tier: quality: "hd" (mj-v8-1) ×1.5, model_params.speed: "turbo" ×2, upscale type: "creative" ×4/3. draft/fast and type: "standard" stay ×1. Measured example: mj-v8-1 + hd + upscale + creative = USD 0.255251 per call.The mj-v7 per-call price moved from USD 0.056723 to USD 0.113446 to match what upstream bills for generation; the action tiers above are unchanged in absolute terms.

Inpaint

Use the mask-image form: model_params.mask.url (an image URL or a data: URL, black = keep, white = regenerate). The polygon form model_params.mask.areas[] is rejected upstream with invalid_parameters: Image dimensions out of range in our measurements (2026-09-13), including with the upstream documentation’s own example coordinates.

Overview

The model page at omnimux.ai/pricing/<model-id> now runs real generations. Its Playground section submits through the gateway using the signed-in browser session, so a user who has already logged in to OmniMux can try a model without creating an API token first.

Which path to call

From the console: the page posts to the session-authenticated POST /pg/video/generations. This path accepts a browser session only and is not part of the token API surface.
From your own code: keep using POST /v1/video/generations with an API token — the same contract documented in the video-series model pages. Nothing about the token path changed.

Task results

Submitted tasks are read back through the same session: GET /api/task/self for status, progress and quota, and GET /api/task/:task_id/artifacts for the media projection. Tasks submitted from the console appear in the account’s task list like any other task.

Billing

Console submissions use the standard precharge and settle path. No separate ledger, pricing rule or quota bucket is introduced for the console path.

Overview

OmniMux now provides Cloudflare R2 object storage integration alongside a local disk fallback driver. File series endpoints are available under /api/v1/files/* and /v1/files/*.

Capabilities

Base64 upload: POST /api/v1/files/upload/base64 supporting Data URL and raw Base64.
Stream upload: POST /api/v1/files/upload/stream via multipart/form-data.
URL upload: POST /api/v1/files/upload/url with SSRF protection.
Quota & Expiration: 2,000 active files per user quota, 72-hour automated expiration with background cleanup.

Links

BreakingImage

Required migration

Replace gpt-image-2 with gpt-image-2.5.
Replace gpt-image-2-hd with gpt-image-2.5-hd.
Keep using POST /v1/images/generations. Update the model value in saved requests, integrations, and any model allowlists.

The old IDs are not compatibility aliases. The gpt-image-2-5 spelling is not supported.

Pricing and scope

Base per-call prices remain USD 0.0441 for gpt-image-2.5 and USD 0.005479 for gpt-image-2.5-hd; group multipliers may affect final charges. This is a public ID migration, not a claim of a newly verified resolution or upstream capability.Flare and Sunburst are separate token-billed models and are not available as part of this change.

Links

GPT Image generation

BreakingVideo

Use operation to select text, first-frame or Max multi-reference generation. Media arrays retain order and conflicting aliases are rejected. Replace legacy 720p with explicit 768P. Temporary result unavailability is retried; failures and cancellations are not masked by stale video URLs. Status reads no longer persist terminal transitions ahead of settlement polling.

fal.ai MiniMax H3 Max & Turbo

MiniMax H3 Max and MiniMax H3 Max Turbo video generation models are now live on OmniMux.

Highlights

Dynamic Multimodal Dispatch: Clients only need to configure two base models (minimax/h3-max and minimax/h3-max-turbo). Passing prompt-only routes to text-to-video, adding image_url routes to image-to-video, and passing video_url routes to reference-to-video automatically.
Ultra-Fast Turbo Performance: Sub-3-second end-to-end rendering for 5-second video clips.
Two-Phase Resilient Polling: Transparent inline result resolution with CDN 404 lag tolerance.
Standard API: Create via POST /v1/video/generations and query via GET /v1/tasks/{task_id}.

Links

BreakingPublishing

The gateway began requiring provider at 2026-09-07 07:25:20 UTC+8 (2026-09-06T23:25:20Z) for listing accounts, starting a connection, disconnecting, creating a post and reading post status. Use tiktok_direct for official TikTok accounts or zernio for Zernio accounts. GET/DELETE carry it in the query; POST carries it in JSON. Missing or unknown values return 400 invalid-provider; a resource from another source returns 409 account-provider-mismatch or 409 post-provider-mismatch. There is no default source.

CLI migration

Use CLI 0.4.0: omnimux update for binaries, or npm install -g @omnimux/cli@0.4.0 for npm. Update all saved commands to include --provider, for example omnimux social accounts --provider zernio. For a read without upstream refresh, use omnimux social post-status <post-id> --provider zernio --no-refresh. Updating a binary does not update your scripts or prove all clients have upgraded.CLI 0.4.0 also includes the previously merged hosted commands and video skill updates. Credentials use a mode-0600 file on every platform; Keychain-only installations need credentials configured securely. The Agent RPC envelope remains contract_version: 1.CLI 0.4.0 was published on 2026-09-07 at 01:47:36 UTC (09:47:36 UTC+8). Both the public binary release and npm package are version 0.4.0.Standalone binaries do not include skill files. Install the npm package to use or update the bundled skills; a binary-only update can report that the skill tree is missing even when the CLI upgrade succeeds.

Links

List accounts

PlatformText

OpenAI Completions, Chat Completions, Responses, and Responses Compact return HTTP 402 with error.type and error.code set to insufficient_quota when the local account balance or subscription allowance cannot cover pre-consume. Streaming requests receive the JSON error before SSE starts. Check balance or allowance before trying again; do not treat this as invalid credentials or temporary rate limiting. Token errors, upstream provider errors, other protocols, actual 401/403, ordinary 429, and 5xx behavior are unchanged.Gateway enforcement began on September 7, 2026 at 07:25:20 UTC+8 (2026-09-06T23:25:20Z).

Language series

Gemini: Added gemini-3.7-flash
GLM: Added glm-5.3
Grok: Added grok-4.6
DeepSeek: Added deepseek-v4-flash-vision-exp

Image & Video series

Canonical model ID hyphenation and full coverage for Wan 3.0 (wan-3.0, wan-3.0-prime, wan-3.0-prime-ref, wan-3.0-ref), Seedance 2.0 (seedance-2-0, seedance-2-0-fast, seedance-2-0-mini, seedance-2-5), Kling (kling-v3, kling-v2-6, kling-avatar, kling-o3, kling-v3-motion-control), Grok Imagine, and Midjourney.
Retired legacy Omni Flash duration-fixed SKUs.

Audio series

gpt-4o-mini-tts (POST /v1/audio/speech)
whisper-1 (POST /v1/audio/transcriptions)
suno (POST /v1/video/generations)

Links

New modelsImageVideo

Image

Model IDs: seedream-5-0-pro, qwen-image-3-0, grok-imagine-image-2-0
API: POST /v1/images/generations

Video

Model IDs: kling-o3, gemini-omni-1.1, pixverse-v6, vidu-q3
API: async POST / GET /v1/video/generations

kling-o3 is a new id and does not replace kling-v3. gemini-omni-1.1 is distinct from gemini-omni-flash. Pin IDs from live pricing.

Jina Reader Relay (`POST /v1/reader`)

Model ID: jina-reader-v1
Capability: URL to clean Markdown / structured JSON text extraction with CSS selectors filtering and output-token billing.
Auth: Standard gateway Bearer sk-... token.

Inspects current active connected social accounts, total quota capacity, and vacant seat availability.

Links

New modelsAudioVideo

New model

Model ID: index-tts
Path: POST / GET /v1/video/generations (GxgenAI task path; result is audio)
Contract: metadata.nodeInfoList required — node 4/audio (reference clip) and 7/text (script)

Docs fix

ltx-2-3-kj is a photo + audio lip-sync model. Prompt-only text-to-video examples were incorrect. Send image node 444 and audio node 1755.

Links

Platform

OmniMux now publishes a dated API Updates feed so you can track gateway model launches and platform changes in one place.

What we log

New models — callable model IDs on live pricing
Model updates — modes, durations, resolutions, parameters
Pricing — material rate changes
Platform — auth, domains, public HTTP surfaces

How to call OmniMux

Surface	URL
Console	`https://omnimux.ai`
OpenAI-compatible API	`https://api.omnimux.ai/v1`
Docs	`https://docs.omnimux.ai`
Status	`https://status.omnimux.ai`

Auth for relay: Bearer sk-… from the console. Full catalog always lives on console pricing — this page is the change timeline.

Links

New modelsSocial data

Social data model IDs are available on OmniMux for read-only public profile and content style access. This is not social publishing.

Model IDs

Platform	IDs
TikTok	`tiktok-user`, `tiktok-posts`, `tiktok-video`, `tiktok-search`
Instagram	`instagram-user`, `instagram-posts`, `instagram-post`, `instagram-search`
X	`x-user`, `x-posts`, `x-tweet`, `x-search`
YouTube	`youtube-user`, `youtube-posts`, `youtube-video`, `youtube-search`

Publishing (different surface)

Connect accounts and create posts use /api/social/v1/* with access token + New-Api-User. Do not mix that auth with gateway sk- Chat Completions.

Links

New modelsVideo

Seedance 2.5

Seedance 2.5 is now available on OmniMux for async video generation.

New model

Model ID: seedance-2-5
Create: POST /v1/video/generations
Poll: GET /v1/video/generations/{task_id}
Highlights: longer single-pass generation and multimodal reference inputs (text / image / video / audio style workflows)

Pin the exact ID from live pricing before integrating.

Links

New modelsVideo

MiniMax H3

MiniMax H3 video models are now available on OmniMux.

New models

Model ID	Role
`minimax-h3`	Standard text/image-to-video
`minimax-h3-t2v`	Text-to-video
`minimax-h3-flf`	First/last-frame style I2V
`minimax-h3-fl2va`	FL2VA-oriented path
`minimax-h3-endframe`	End-frame reference I2V

API

Create: POST /v1/video/generations
Poll: GET /v1/video/generations/{task_id}

Async only — submit once, then poll task status.

Links

New modelsVideo

Grok Imagine Video 1.5

Grok Imagine Video 1.5 is now available on OmniMux.

New model

Model ID: grok-imagine-video-1-5
Mode: image-to-video style generation via async video API
Create / poll: POST / GET /v1/video/generations

See the model page for parameters and the task poll contract.

Links

New modelsText

Claude Opus 5

Claude Opus 5 is now available on OmniMux through OpenAI-compatible Chat Completions.

New model

Model ID: claude-opus-5
Endpoint: https://api.omnimux.ai/v1 (Chat Completions)
Docs: one complete page per brand — model is an enum on the Claude complete page

Related Claude IDs on the gateway include other Opus / Sonnet / Haiku SKUs on live pricing.

Gemini Flash models

New Gemini Flash-class models are available on OmniMux.

New models

Model IDs: gemini-3.6-flash, gemini-3.5-flash
API: OpenAI-compatible Chat Completions at https://api.omnimux.ai/v1

Additional Gemini IDs (Pro / preview / lite) may also appear on live pricing — pin the exact ID from the console or Gemini complete page.

Kimi K3

Kimi K3 is now available on OmniMux.

New model

Model ID: kimi-k3
API: Chat Completions at https://api.omnimux.ai/v1
Focus: long-context reasoning and agent-style workflows

See the Kimi complete page for the full model enum.

Suno

Suno music generation is now available on OmniMux.

New model

Model ID: suno
Use: AI music generation workflows via the gateway catalog

Confirm billing and request shape from live pricing / console before production traffic.

Nano Banana

Nano Banana 2 and Nano Banana Pro are now available on OmniMux.

New models

Model IDs: nano-banana-2, nano-banana-pro
Modality: image generation

See each model page for request parameters.

Links

New modelsVideo

Seedance 2.0

The Seedance 2.0 family is available on OmniMux for async video generation.

New models

Model ID	Role
`seedance-2-0`	Standard
`seedance-2-0-mini`	Lightweight / cost-oriented
`seedance-2-0-fast`	Speed-oriented

API

Create: POST /v1/video/generations
Poll: GET /v1/video/generations/{task_id}

Links

New modelsText

GLM-5.2

GLM-5.2 is now available on OmniMux (with related GLM IDs on pricing).

New models

Model IDs: glm-5.2, glm-5.1
API: Chat Completions at https://api.omnimux.ai/v1

Kling

Kling video models are available on OmniMux.

New models

Model IDs: kling-v3, kling-v2-6
API: async POST / GET /v1/video/generations

Pin IDs from live pricing; request parameters follow the video task contract.

Links

New modelsImage

Midjourney

Midjourney is available on OmniMux as model ID midjourney.

New model

Model ID: midjourney
Modality: image generation via the gateway catalog

Confirm request shape and billing on console pricing before production use.

MiniMax M3

MiniMax M3 is now available on OmniMux Chat Completions.

New model

Model ID: minimax-m3
API: https://api.omnimux.ai/v1

Omni Flash

Omni Flash (gemini-omni-flash) is available on OmniMux for fast video generation.

Model

Model ID: gemini-omni-flash
API: POST /v1/video/generations + poll GET /v1/video/generations/{task_id}

Links

New modelsVideo

Veo 3.1

Veo 3.1 is now available on OmniMux.

New model

Model ID: veo-3.1
API: async video create + poll on /v1/video/generations

Links

New modelsText

Claude Opus 4.7 and DeepSeek V4

New language models are available on OmniMux Chat Completions.

New models

Model ID	Brand
`claude-opus-4-7`	Claude
`deepseek-v4-flash`	DeepSeek
`deepseek-v4-pro`	DeepSeek

API

Endpoint: https://api.omnimux.ai/v1 (Chat Completions)

Links

New modelsImage

GPT Image 2

GPT Image 2 is now available on OmniMux.

New models

Model IDs: gpt-image2, gpt-image2-hd
Modality: image generation

See model docs for size / quality parameters.

Links

New modelsText

MiniMax M2.5

MiniMax M2.5 (and related MiniMax text IDs) are available on OmniMux.

New models

Model IDs: minimax-m2.5, minimax-m2.7
API: Chat Completions

GPT-5.4

GPT-5.4 series models are available on OmniMux Chat Completions.

New models

Model IDs: gpt-5.4, gpt-5.4-mini
API: https://api.omnimux.ai/v1

Later GPT-5.x IDs (for example gpt-5.5, gpt-5.6-terra) may also appear on live pricing — always pin the ID from the console.

​GPT Image Tiers

​Model specification

​Request example

​Links

​Video Task and Variant Models

​Model specification

​Request example

​Links

​Doubao Speech Recognition

​Model specification

​Request example

​Links

​Seed Audio 1.0

​Model specification

​Request example

​Links

​Index TTS 2

​Model specification

​Request example

​Links

​Suno Sound Effects

​Model specification

​Request example

​Links

​Gemini 3.8 Flash

​Model specification

​Request example

​Links

​What is new

​Documentation

​Links

​What changed

​Limits we do not state

​Withdrawn figure

​Documentation

​Links

​Required migration

​Pricing

​Inpaint

​Links

​Overview

​Which path to call

​Task results

​Billing

​Links

​Overview

​Capabilities

​Links

​Required migration

​Pricing and scope

​Links

​Links

​fal.ai MiniMax H3 Max & Turbo

​Highlights

​Links

​CLI migration

​Links

​Links

​Language series

​Image & Video series

​Audio series

​Links

​Image

​Video

​Links

​Social Analytics & Inbox Suite

​Jina Reader Relay (POST /v1/reader)

​Social Seats Management (GET /api/social/v1/seats)

​Links

​New model

​Docs fix

​Links

​API Updates

​What we log

​How to call OmniMux

​Links

​Social data (read-only)

​Model IDs

​Publishing (different surface)

​Links

GPT Image Tiers

Model specification

Request example

Links

Video Task and Variant Models

Model specification

Request example

Links

Doubao Speech Recognition

Model specification

Request example

Links

Seed Audio 1.0

Model specification

Request example

Links

Index TTS 2

Model specification

Request example

Links

Suno Sound Effects

Model specification

Request example

Links

Gemini 3.8 Flash

Model specification

Request example

Links

What is new

Documentation

Links

What changed

Limits we do not state

Withdrawn figure

Documentation

Links

Required migration

Pricing

Inpaint

Links

Overview

Which path to call

Task results

Billing

Links

Overview

Capabilities

Links

Required migration

Pricing and scope

Links

Links

fal.ai MiniMax H3 Max & Turbo

Highlights

Links

CLI migration

Links

Links

Language series

Image & Video series

Audio series

Links

Image

Video

Links

Social Analytics & Inbox Suite

Jina Reader Relay (`POST /v1/reader`)

Social Seats Management (`GET /api/social/v1/seats`)

Links

New model

Docs fix

Links

API Updates

What we log

How to call OmniMux

Links

Social data (read-only)

Model IDs

Publishing (different surface)

Links