# Aiduct

> Aiduct is an AI API gateway for developers building with chat, image, video,
> audio and embedding models. Use one API key and a prepaid USD balance.

OpenAI-compatible endpoints, native Anthropic Messages and OpenAI Responses
are available. Compatibility, parameters, billing units and availability vary
by model; consult the model catalog and endpoint documentation for each request.

## Primary URLs

- Website: https://aiduct.ai
- API base URL: https://api.aiduct.ai/v1
- Docs: https://aiduct.ai/docs
- Models: https://aiduct.ai/models
- Pricing: https://aiduct.ai/pricing
- Playground: https://aiduct.ai/playground
- Blog: https://aiduct.ai/blog
- Sitemap: https://aiduct.ai/sitemap.xml
- Full text reference: https://aiduct.ai/llms-full.txt
- OpenAPI specification: https://api.aiduct.ai/openapi.json
- Live model catalog and USD price schedules: https://api.aiduct.ai/v1/models/catalog

## Coding agents and SDKs

- Claude Code: https://aiduct.ai/for/claude-code — set `ANTHROPIC_BASE_URL` to
  `https://api.aiduct.ai`; the native Anthropic Messages protocol is served
  directly, not translated through an OpenAI shape.
- Codex CLI: https://aiduct.ai/for/codex — add a model provider in
  `~/.codex/config.toml` with `base_url = "https://api.aiduct.ai/v1"` and
  `wire_api = "responses"`; the OpenAI Responses protocol is served for every
  chat model.
- Gemini CLI: https://aiduct.ai/for/gemini-cli — set `GOOGLE_GEMINI_BASE_URL` to
  `https://api.aiduct.ai` and `GEMINI_API_KEY` to the Aiduct key; the Gemini
  `generateContent` API is served at `/v1beta`.
- pi coding agent: https://aiduct.ai/for/pi — add Aiduct providers to
  `~/.pi/agent/models.json`; every API pi supports (`openai-completions`,
  `openai-responses`, `anthropic-messages`, `google-generative-ai`) works.
- OpenAI SDK: https://aiduct.ai/for/openai-sdk — point `base_url` at
  `https://api.aiduct.ai/v1` and keep the existing client code.
- Every chat model answers on every chat protocol: natively where its upstream
  speaks the protocol, translated otherwise.
- OpenRouter comparison: https://aiduct.ai/alternatives/openrouter

## Endpoints

- `POST /v1/chat/completions` — OpenAI-compatible chat, streaming supported;
  `usage.prompt_tokens` counts cached input, with `cached_tokens` and
  `cache_write_tokens` as parts of it.
- `POST /v1/responses` and `POST /v1/responses/input_tokens` — OpenAI Responses
  protocol and its input-token count.
- `POST /v1/messages` and `POST /v1/messages/count_tokens` — Anthropic Messages
  protocol and its token count.
- `POST /v1beta/models/{model}:generateContent`, `:streamGenerateContent` and
  `:countTokens` — Google Gemini protocol; `GET /v1beta/models` lists models.
- `POST /v1/images/generations` — image models.
- `POST /v1/videos/generations` — video generation; use the queue endpoints for asynchronous jobs.
- `POST /v1/embeddings` — embedding models.
- `POST /v1/audio/speech` — audio generation; parameters depend on the model.
- `POST /v1/audio/transcriptions` — speech transcription with multipart file upload.
- `GET /v1/models` — model catalog with pricing metadata.
- Full reference: https://aiduct.ai/docs

## Billing

Prepaid balance in USD through Stripe. Every request records usage against the
API key; the dashboard links each charge to the activity that produced it.
There is no monthly subscription and no markup on credit purchases.
Rates are published per input/output token, image, video second, audio unit,
request or another explicitly named unit. Different units must not be compared
as an undifferentiated price range. Some models have additional option-dependent
charges. Read the live catalog before quoting a price or estimating a bill.

## Policies and support

- Terms: https://aiduct.ai/terms
- Privacy: https://aiduct.ai/privacy
- Billing and refunds: https://aiduct.ai/refunds

## Contact

hello@aiduct.ai

## Reference snapshot

Catalog timestamp: 2026-10-23T03:08:04.339Z. This is a build-time reference, not a real-time availability or price guarantee. The live catalog at https://api.aiduct.ai/v1/models/catalog is authoritative; rates exclude any separately listed model-option charges.

## API endpoint reference

### POST /v1/chat/completions

Create a chat completion

OpenAI ChatCompletion-compatible endpoint. Set `stream: true` to receive a Server-Sent Events stream where each `data:` line is a `chat.completion.chunk` payload terminated by `data: [DONE]`. Model id is the API-platform id from `GET /v1/models`.

Reference: https://aiduct.ai/docs/create-chat-completion

Example request body:

```json
{
  "model": "claude-sonnet-4-5",
  "messages": [
    {
      "role": "user",
      "content": "Hello!"
    }
  ]
}
```

### POST /v1/messages

Anthropic Messages (Claude Code protocol)

Anthropic's Messages protocol, for Claude Code (`ANTHROPIC_BASE_URL=https://api.aiduct.ai`, no `/v1`) and the Anthropic SDKs. Every catalog chat model answers here: Claude models natively, others translated, with tool use, thinking and images kept. Authenticate with `x-api-key` or `Authorization: Bearer`; `anthropic-version` and `anthropic-beta` reach a natively served model. Set `stream: true` for Anthropic's `event: <type>` SSE stream. Errors use Anthropic's envelope. `usage` counts `input_tokens` without the cache, beside `cache_read_input_tokens` and `cache_creation_input_tokens`.

Reference: https://aiduct.ai/docs/create-anthropic-message

Example request body:

```json
{
  "model": "claude-sonnet-4-5",
  "max_tokens": 1024,
  "messages": [
    {
      "role": "user",
      "content": "Hi"
    }
  ]
}
```

### POST /v1/messages/count_tokens

Count the input tokens of a Messages request

Anthropic's token-counting endpoint. A model served natively on this protocol is counted by its upstream; any other count is an estimate, marked by `x-aiduct-token-count: estimated`. Not billed.

Reference: https://aiduct.ai/docs/count-anthropic-message-tokens

### POST /v1/responses

OpenAI Responses API (Codex CLI protocol)

OpenAI's Responses protocol, for the Codex CLI (a model provider with `base_url = "https://api.aiduct.ai/v1"` and `wire_api = "responses"`) and `responses.create` in the OpenAI SDKs. Every catalog chat model answers here: OpenAI models natively where the route allows it, others translated, including Codex's function, custom (`apply_patch`) and local shell tools and encrypted reasoning round trips. Stateless: send the whole conversation each turn. Set `stream: true` for the Responses SSE event stream.

Reference: https://aiduct.ai/docs/create-open-a-i-response

Example request body:

```json
{
  "model": "gpt-4o",
  "input": "Write a haiku about API gateways."
}
```

### POST /v1/responses/input_tokens

Count the input tokens of a Responses request

OpenAI's input-token counting endpoint. A model served natively on this protocol is counted by its upstream; any other count is an estimate, marked by `x-aiduct-token-count: estimated`. Not billed.

Reference: https://aiduct.ai/docs/count-open-a-i-response-input-tokens

### POST /v1beta/models/{model}:generateContent

Gemini generateContent

Google's Gemini protocol, for the Gemini CLI (`GOOGLE_GEMINI_BASE_URL=https://api.aiduct.ai`), `@google/genai` (`httpOptions.baseUrl`) and pi. Every catalog chat model answers here: Gemini models natively, others translated, with function calling and thinking kept. Authenticate with `x-goog-api-key`. Errors use Google's envelope.

Reference: https://aiduct.ai/docs/gemini-generate-content

### POST /v1beta/models/{model}:streamGenerateContent

Gemini streamGenerateContent

The streamed form of `generateContent`. Send `?alt=sse` for a server-sent event per `GenerateContentResponse` chunk, as the Gemini SDKs do.

Reference: https://aiduct.ai/docs/gemini-stream-generate-content

### POST /v1beta/models/{model}:countTokens

Gemini countTokens

Gemini's token-counting endpoint. A model served natively on this protocol is counted by its upstream; any other count is an estimate, marked by `x-aiduct-token-count: estimated`. Not billed.

Reference: https://aiduct.ai/docs/gemini-count-tokens

### POST /v1/images/generations

Generate or edit an image from a prompt

OpenAI Images-compatible endpoint. Some models, such as `gpt-image-2`, also accept `input_media` reference images. No streaming.

Reference: https://aiduct.ai/docs/create-image-generation

Example request body:

```json
{
  "model": "flux-schnell",
  "prompt": "a red panda eating bamboo, studio photo",
  "size": "1024x1024"
}
```

### POST /v1/videos/generations

Submit an async video-generation job

Always async. Returns a queue ticket; poll the `status_url` until the job completes, then read the final result from `response_url`. Use `cancel_url` to abort. Optional `webhook_url` lets us POST `response.completed` to a customer endpoint (validated against SSRF: must be https in production, must resolve to a public IP).

Reference: https://aiduct.ai/docs/create-video-generation

Example request body:

```json
{
  "model": "kling-v2.1",
  "prompt": "a red panda eating bamboo, cinematic",
  "duration": 5,
  "aspect_ratio": "16:9"
}
```

### GET /v1/queue/requests/{id}

Fetch the queued job's result (or terminal-state payload)

200 with the final result when status is `completed`. 200 with `{status:'failed'|'cancelled', error, ...}` when terminal but unsuccessful (no body retry, just inspect). 202 with the polling payload when still in progress.

Reference: https://aiduct.ai/docs/get-queue-request-result

### GET /v1/queue/requests/{id}/status

Cheap poll of a job's current state

Always returns 200 with the latest status. Safe to call at sub-second cadence — does not touch upstream providers.

Reference: https://aiduct.ai/docs/get-queue-request-status

### POST /v1/queue/requests/{id}/cancel

Cancel a queued job

Marks the job as cancelled and best-effort asks the upstream provider to stop. Does NOT refund compute already consumed. Returns 409 if the job is already in a terminal state.

Reference: https://aiduct.ai/docs/cancel-queue-request

### POST /v1/embeddings

Create vector embeddings for text input

OpenAI-compatible embeddings endpoint. Pass a single string or an array; receive one float (or base64) vector per input. Use `text-embedding-3-small` for cheap baselines and `text-embedding-3-large` for higher retrieval quality.

Reference: https://aiduct.ai/docs/create-embedding

Example request body:

```json
{
  "model": "text-embedding-3-small",
  "input": "Aiduct is a unified AI API gateway."
}
```

### POST /v1/audio/speech

Generate speech or queued music from text

Model-specific audio generation endpoint. Speech models return audio bytes synchronously. `minimax-music-v2` returns a 202 queue ticket and rejects speech-only fields such as `voice` and `speed`.

Reference: https://aiduct.ai/docs/create-audio-speech

Example request body:

```json
{
  "model": "tts-1",
  "input": "Hello from Aiduct.",
  "voice": "alloy"
}
```

### POST /v1/audio/transcriptions

Transcribe audio into text

OpenAI-compatible multipart speech-to-text endpoint. The gateway returns JSON transcription responses.

Reference: https://aiduct.ai/docs/create-audio-transcription

### GET /v1/models

OpenAI-compatible model list

Returns every model the gateway can route to in OpenAI's `{object:'list', data:[…]}` shape, augmented with `available`, `display_name`, `description`, and `pricing`. A request carrying `anthropic-version`, as the Anthropic SDKs send, gets the chat models in Anthropic's `{data, has_more, first_id, last_id}` shape instead.

Reference: https://aiduct.ai/docs/list-models

### GET /v1/models/{model}

Retrieve one model

One model by its catalog id or a declared alias, in OpenAI's shape, or Anthropic's when the request carries `anthropic-version`.

Reference: https://aiduct.ai/docs/retrieve-model

### GET /v1beta/models

Gemini models.list

The chat models as Gemini `Model` resources named `models/{id}`, each supporting `generateContent`, `streamGenerateContent` and `countTokens`.

Reference: https://aiduct.ai/docs/gemini-list-models

### GET /v1beta/models/{model}

Gemini models.get

One chat model as a Gemini `Model` resource.

Reference: https://aiduct.ai/docs/gemini-get-model

### GET /v1/models/catalog

SEO-friendly merged model catalog

Slug-friendly catalog used by aiduct.ai marketing pages. Includes pre-formatted price labels and a single-line price summary. Cached 60s.

Reference: https://aiduct.ai/docs/get-public-model-catalog

## Models with published rates

### Flux Pro 1.1

- Model ID: `flux-pro-v1.1`
- Model page: https://aiduct.ai/models/flux-pro-v1.1
- Provider: vercel
- Capabilities: image-generation, image-editing
- Available at snapshot: yes

High-quality professional image generation

- Images: USD 0.066 per 1 image

### ChatGPT Image 2

- Model ID: `gpt-image-2`
- Model page: https://aiduct.ai/models/gpt-image-2
- Provider: vercel
- Capabilities: image-generation, image-editing
- Available at snapshot: yes

OpenAI GPT Image 2 generation and reference-image editing via OpenAI official, with Vercel AI Gateway and fal.ai fallbacks

- Images: USD 0.087 per 1 image

### ChatGPT Image 2.5 Flare

- Model ID: `gpt-image-2.5-flare`
- Model page: https://aiduct.ai/models/gpt-image-2.5-flare
- Provider: vercel
- Capabilities: image-generation, image-editing
- Available at snapshot: yes

Fast everyday image generation and reference editing. Uses the exact OpenAI model through Vercel AI Gateway; billed at reported upstream cost.

- Input tokens: USD 5 per 1,000,000 input tokens
- Output tokens: USD 30 per 1,000,000 output tokens
- Cached input: USD 1.25 per 1,000,000 cached input tokens

### ChatGPT Image 2.5 Sunburst

- Model ID: `gpt-image-2.5-sunburst`
- Model page: https://aiduct.ai/models/gpt-image-2.5-sunburst
- Provider: vercel
- Capabilities: image-generation, image-editing
- Available at snapshot: yes

Precise image generation and reference editing. Uses the exact OpenAI model through Vercel AI Gateway; billed at reported upstream cost.

- Input tokens: USD 5 per 1,000,000 input tokens
- Output tokens: USD 30 per 1,000,000 output tokens
- Cached input: USD 1.25 per 1,000,000 cached input tokens

### Hailuo 2.3

- Model ID: `hailuo-2.3`
- Model page: https://aiduct.ai/models/hailuo-2.3
- Provider: minimax
- Capabilities: video-generation
- Available at snapshot: yes

Strong motion model for action-heavy scenes, fights, and kinetic camera work

- Video duration: USD 0.072 per 1 video second

### MiniMax H3 Max

- Model ID: `hailuo-3-max`
- Model page: https://aiduct.ai/models/hailuo-3-max
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

fal's post-trained MiniMax H3 variant for stronger prompt adherence, cinematic aesthetics, and faster-than-realtime generation

- Video duration: USD 0.08 per 1 video second

### HunyuanVideo

- Model ID: `hunyuan-video`
- Model page: https://aiduct.ai/models/hunyuan-video
- Provider: fal
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Tencent HunyuanVideo: open-source 5-second text and image-to-video generation.

- Video duration: USD 0.081 per 1 video second

### Ideogram V3

- Model ID: `ideogram-v3`
- Model page: https://aiduct.ai/models/ideogram-v3
- Provider: fal
- Capabilities: image-generation, image-editing
- Available at snapshot: yes

Excellent at text rendering and typography in images

- Images: USD 0.096 per 1 image

### Kling Image 1.0

- Model ID: `kling-image-v1`
- Model page: https://aiduct.ai/models/kling-image-v1
- Provider: kling
- Capabilities: image-generation, image-editing
- Available at snapshot: yes

Kolors 1.0 text-to-image generation

- Images: USD 0.004 per 1 image

### DeepSeek V4 Flash 0731

- Model ID: `deepseek-v4-flash`
- Model page: https://aiduct.ai/models/deepseek-v4-flash
- Provider: zenmux
- Capabilities: chat, tools, thinking, structured-output
- Available at snapshot: yes

DeepSeek's July 31 V4 Flash release for long-context reasoning, structured output, and agent tool use.

- Input tokens: USD 0.44 per 1,000,000 input tokens
- Output tokens: USD 1.32 per 1,000,000 output tokens
- Cached input: USD 0.014 per 1,000,000 cached input tokens

### DeepSeek V4.1 Flash

- Model ID: `deepseek-v4.1-flash`
- Model page: https://aiduct.ai/models/deepseek-v4.1-flash
- Provider: zenmux
- Capabilities: chat, vision, tools, thinking, structured-output
- Available at snapshot: yes

DeepSeek's September 10 V4.1 Flash release with image understanding, long-context reasoning, JSON output, and agent tool use.

- Input tokens: USD 0.3 per 1,000,000 input tokens
- Output tokens: USD 1.2 per 1,000,000 output tokens
- Cached input: USD 0.006 per 1,000,000 cached input tokens

### DeepSeek V4 Pro 0813

- Model ID: `deepseek-v4-pro`
- Model page: https://aiduct.ai/models/deepseek-v4-pro
- Provider: zenmux
- Capabilities: chat, tools, thinking, structured-output
- Available at snapshot: yes

DeepSeek's August 13 V4 Pro release for long-context reasoning, structured output, and agent tool use.

- Input tokens: USD 1.32 per 1,000,000 input tokens
- Output tokens: USD 3.9600000000000004 per 1,000,000 output tokens
- Cached input: USD 0.044 per 1,000,000 cached input tokens

### Kling Image 1.5

- Model ID: `kling-image-v1.5`
- Model page: https://aiduct.ai/models/kling-image-v1.5
- Provider: kling
- Capabilities: image-generation, image-editing
- Available at snapshot: yes

Kolors 1.5 text-to-image generation

- Images: USD 0.014 per 1 image

### Kling Image 2.0

- Model ID: `kling-image-v2`
- Model page: https://aiduct.ai/models/kling-image-v2
- Provider: kling
- Capabilities: image-generation, image-editing
- Available at snapshot: yes

Kolors 2.0 high-fidelity text-to-image generation

- Images: USD 0.028 per 1 image

### Kling 1.6

- Model ID: `kling-v1.6`
- Model page: https://aiduct.ai/models/kling-v1.6
- Provider: kling
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Kling 1.6 Pro cinematic video generation

- Video duration: USD 0.098 per 1 video second

### Kling 1.6 Standard

- Model ID: `kling-v1.6-standard`
- Model page: https://aiduct.ai/models/kling-v1.6-standard
- Provider: kling
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Kling 1.6 Standard cost-efficient video generation

- Video duration: USD 0.049 per 1 video second

### Kling 2.0 Master

- Model ID: `kling-v2-master`
- Model page: https://aiduct.ai/models/kling-v2-master
- Provider: kling
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Kling 2.0 Master premium cinematic video generation

- Video duration: USD 0.28 per 1 video second

### Kling 2.1

- Model ID: `kling-v2.1`
- Model page: https://aiduct.ai/models/kling-v2.1
- Provider: kling
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

High-quality cinematic video generation

- Video duration: USD 0.159 per 1 video second

### Kling 2.1 Standard

- Model ID: `kling-v2.1-standard`
- Model page: https://aiduct.ai/models/kling-v2.1-standard
- Provider: kling
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Kling 2.1 standard tier for lower-cost image-to-video generation.

- Video duration: USD 0.055 per 1 video second

### Gemini 3.8 Flash

- Model ID: `gemini-3.8-flash`
- Model page: https://aiduct.ai/models/gemini-3.8-flash
- Provider: google
- Capabilities: chat, vision, tools, thinking, structured-output
- Available at snapshot: yes

Google's newest Flash model for fast agentic workflows, long-context multimodal reasoning, and tool use.

- Input tokens: USD 1.5 per 1,000,000 input tokens
- Output tokens: USD 7.5 per 1,000,000 output tokens
- Cached input: USD 0.15 per 1,000,000 cached input tokens

### Gemini 3.7 Flash

- Model ID: `gemini-3.7-flash`
- Model page: https://aiduct.ai/models/gemini-3.7-flash
- Provider: google
- Capabilities: chat, vision, tools, thinking, structured-output
- Available at snapshot: yes

Google's most capable Flash model for agentic workflows, long-context multimodal reasoning, and tool use.

- Input tokens: USD 1.5 per 1,000,000 input tokens
- Output tokens: USD 7.5 per 1,000,000 output tokens
- Cached input: USD 0.15 per 1,000,000 cached input tokens

### Gemini 3.5 Flash

- Model ID: `gemini-3.5-flash`
- Model page: https://aiduct.ai/models/gemini-3.5-flash
- Provider: google
- Capabilities: chat, vision, tools, thinking, structured-output
- Available at snapshot: yes

Google's latest agentic Flash model for coding, tool use, long-context reasoning, and multimodal chat.

- Input tokens: USD 1.5 per 1,000,000 input tokens
- Output tokens: USD 9 per 1,000,000 output tokens
- Cached input: USD 0.15 per 1,000,000 cached input tokens

### Kling 2.5 Turbo

- Model ID: `kling-v2.5-turbo`
- Model page: https://aiduct.ai/models/kling-v2.5-turbo
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Kling 2.5 Turbo Pro for fast text and image-to-video with first/last frames.

- Video duration: USD 0.114 per 1 video second

### Kling 2.5 Turbo Standard

- Model ID: `kling-v2.5-turbo-standard`
- Model page: https://aiduct.ai/models/kling-v2.5-turbo-standard
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Kling 2.5 Turbo standard image-to-video oriented tier for fast motion drafts and short clips.

- Video duration: USD 0.06 per 1 video second

### GPT-4o

- Model ID: `gpt-4o`
- Model page: https://aiduct.ai/models/gpt-4o
- Provider: openai
- Capabilities: chat, vision, tools, structured-output
- Available at snapshot: yes

OpenAI's flagship multimodal model. Strong reasoning, tool use, and vision; the right default when GPT-4o-mini is not capable enough.

- Input tokens: USD 2.5 per 1,000,000 input tokens
- Output tokens: USD 10 per 1,000,000 output tokens

### GPT-4o mini

- Model ID: `gpt-4o-mini`
- Model page: https://aiduct.ai/models/gpt-4o-mini
- Provider: openai
- Capabilities: chat, vision, tools, structured-output
- Available at snapshot: yes

OpenAI's small, fast, cheap multimodal model. The default choice for routine chat, structured output, and short coding tasks.

- Input tokens: USD 0.15 per 1,000,000 input tokens
- Output tokens: USD 0.6 per 1,000,000 output tokens

### Kling 2.6 Pro

- Model ID: `kling-v2.6-pro`
- Model page: https://aiduct.ai/models/kling-v2.6-pro
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Kling 2.6 Pro text and image-to-video tier with cinematic motion and native-audio capable upstream support.

- Video duration: USD 0.084 per 1 video second

### Kling V3.0

- Model ID: `kling-v3`
- Model page: https://aiduct.ai/models/kling-v3
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Kling 3.0 Pro with cinematic motion, native audio, and 3-15 second clips

- Video duration: USD 0.159 per 1 video second

### Kling 3.0 Standard

- Model ID: `kling-v3-standard`
- Model page: https://aiduct.ai/models/kling-v3-standard
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Kling 3.0 standard tier for cinematic text and image-to-video generation.

- Video duration: USD 0.105 per 1 video second

### Grok 4.6

- Model ID: `grok-4.6`
- Model page: https://aiduct.ai/models/grok-4.6
- Provider: zenmux
- Capabilities: chat, vision, tools, thinking, structured-output
- Available at snapshot: yes

xAI's Grok 4.6 for agent tool use, long-context reasoning, and multimodal input.

- Input tokens: USD 2 per 1,000,000 input tokens
- Output tokens: USD 6 per 1,000,000 output tokens
- Cached input: USD 0.5 per 1,000,000 cached input tokens

### Luma Ray2

- Model ID: `luma-ray2`
- Model page: https://aiduct.ai/models/luma-ray2
- Provider: fal
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

High-quality creative video with loop support

- Video duration: USD 0.162 per 1 video second

### Luma Ray2 Flash

- Model ID: `luma-ray2-flash`
- Model page: https://aiduct.ai/models/luma-ray2-flash
- Provider: fal
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Fast variant of Ray2 for rapid video iteration

- Video duration: USD 0.066 per 1 video second

### MiniMax H3

- Model ID: `minimax-h3`
- Model page: https://aiduct.ai/models/minimax-h3
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

MiniMax's latest native multimodal video model with 2K output, first/last-frame control, and image, video, and audio references

- Video duration: USD 0.13 per 1 video second

### MiniMax Video-01

- Model ID: `minimax-video-01`
- Model page: https://aiduct.ai/models/minimax-video-01
- Provider: minimax
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

MiniMax Video-01: fixed 6-second 720p text or image-to-video with prompt optimization.

- Requests: USD 0.801 per 1 request

### Nano Banana

- Model ID: `nano-banana`
- Model page: https://aiduct.ai/models/nano-banana
- Provider: vercel
- Capabilities: image-generation, image-editing
- Available at snapshot: yes

Original Nano Banana for rapid iteration at low cost

- Images: USD 0.063 per 1 image

### Nano Banana 2

- Model ID: `nano-banana-2`
- Model page: https://aiduct.ai/models/nano-banana-2
- Provider: vercel
- Capabilities: image-generation, image-editing
- Available at snapshot: yes

Fast reasoning-guided image generation with text rendering and character consistency

- Images: USD 0.108 per 1 image

### Nano Banana Pro

- Model ID: `nano-banana-pro`
- Model page: https://aiduct.ai/models/nano-banana-pro
- Provider: vercel
- Capabilities: image-generation, image-editing
- Available at snapshot: yes

Highest quality image generation with deep reasoning and up to 4K resolution

- Images: USD 0.24 per 1 image

### PixVerse v4

- Model ID: `pixverse-v4`
- Model page: https://aiduct.ai/models/pixverse-v4
- Provider: fal
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

PixVerse v4 — previous generation, lower cost than v4.5.

- Video duration: USD 0.15 per 1 video second

### PixVerse v4.5

- Model ID: `pixverse-v4.5`
- Model page: https://aiduct.ai/models/pixverse-v4.5
- Provider: fal
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

PixVerse v4.5 — stylized video generation with strong anime/stylized motion.

- Video duration: USD 0.25 per 1 video second

### Seedance 2.0

- Model ID: `seedance-2.0`
- Model page: https://aiduct.ai/models/seedance-2.0
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Most capable general-purpose video model with multimodal image, video, and audio references

- Video duration: USD 0.486 per 1 video second

### Seedance 2.0 Fast

- Model ID: `seedance-2.0-fast`
- Model page: https://aiduct.ai/models/seedance-2.0-fast
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Lower-latency, lower-cost Seedance 2.0 variant with multimodal reference support

- Video duration: USD 0.39 per 1 video second

### Seedance 2.0 Fast Reference

- Model ID: `seedance-2.0-fast-reference`
- Model page: https://aiduct.ai/models/seedance-2.0-fast-reference
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Lower-latency Seedance 2.0 multimodal reference-to-video workflow for image, video, and audio guided short-form generation.

- Video duration: USD 0.39 per 1 video second

### Seedance 2.0 Mini

- Model ID: `seedance-2.0-mini`
- Model page: https://aiduct.ai/models/seedance-2.0-mini
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Cost-efficient Seedance 2.0 variant with multimodal image, video, and audio references

- Video duration: USD 0.39 per 1 video second

### Seedance 2.0 Reference

- Model ID: `seedance-2.0-reference`
- Model page: https://aiduct.ai/models/seedance-2.0-reference
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Seedance 2.0 multimodal reference-to-video workflow for image, video, and audio guided character, motion, scene, and rhythm consistency.

- Video duration: USD 0.486 per 1 video second

### Google Lyria 3

- Model ID: `lyria-3`
- Model page: https://aiduct.ai/models/lyria-3
- Provider: fal
- Capabilities: music-generation
- Available at snapshot: yes

Multilingual music with vocals, lyrics, or instrumental arrangements described in the input. MP3 output through the asynchronous audio queue.

- Requests: USD 0.04 per 1 request

### Seedance 2.5

- Model ID: `seedance-2.5`
- Model page: https://aiduct.ai/models/seedance-2.5
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

ByteDance's latest professional multimodal video model with up to 30-second native-audio generation and stronger reference control

- Video duration: USD 0.486 per 1 video second

### Google Lyria 3 Pro

- Model ID: `lyria-3-pro`
- Model page: https://aiduct.ai/models/lyria-3-pro
- Provider: fal
- Capabilities: music-generation
- Available at snapshot: yes

High-quality multilingual songs and arrangements with vocals and lyrics described in the input. MP3 output through the asynchronous audio queue.

- Requests: USD 0.08 per 1 request

### ElevenLabs Music (60s)

- Model ID: `elevenlabs-music`
- Model page: https://aiduct.ai/models/elevenlabs-music
- Provider: fal
- Capabilities: music-generation
- Available at snapshot: yes

Generate a 60-second song or instrumental track from a text description. This platform model fixes the duration to 60 seconds and the output to MP3; each completed track is billed once. Returns an asynchronous audio polling ticket.

- Requests: USD 0.6 per 1 request

### Seedance 1.0 Pro

- Model ID: `seedance-v1.0-pro`
- Model page: https://aiduct.ai/models/seedance-v1.0-pro
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Official Seedance 1.0 quality model for text, first-frame, and first/last-frame video generation

- Video duration: USD 0.36 per 1 video second

### Seedance 1.0 Pro Fast

- Model ID: `seedance-v1.0-pro-fast`
- Model page: https://aiduct.ai/models/seedance-v1.0-pro-fast
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Official lower-latency Seedance 1.0 variant for text and first-frame video generation

- Video duration: USD 0.1 per 1 video second

### Seedance 1.5 Pro

- Model ID: `seedance-v1.5-pro`
- Model page: https://aiduct.ai/models/seedance-v1.5-pro
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Official Seedance 1.5 Pro with native audio and first/last-frame control; scheduled for Volcengine EOS on 2026-09-21

- Video duration: USD 0.129 per 1 video second

### Seedream 4.0

- Model ID: `seedream-v4`
- Model page: https://aiduct.ai/models/seedream-v4
- Provider: vercel
- Capabilities: image-generation, image-editing
- Available at snapshot: yes

ByteDance Seedream 4.0 text-to-image model with strong text rendering, composition, and commercial image workflows.

- Images: USD 0.04 per 1 image

### Seedream 4.5

- Model ID: `seedream-v4.5`
- Model page: https://aiduct.ai/models/seedream-v4.5
- Provider: vercel
- Capabilities: image-generation, image-editing
- Available at snapshot: yes

ByteDance Seedream 4.5 text-to-image model for high-resolution commercial image generation and prompt adherence.

- Images: USD 0.045 per 1 image

### Sora 2

- Model ID: `sora-2`
- Model page: https://aiduct.ai/models/sora-2
- Provider: fal
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

OpenAI's flagship video generation model. Strong physics, instruction following, and synchronized audio. Routed via fal.ai until the OpenAI provider gains a video adapter.

- Video duration: USD 0.15 per 1 video second

### o3-mini

- Model ID: `o3-mini`
- Model page: https://aiduct.ai/models/o3-mini
- Provider: openai
- Capabilities: chat, tools, structured-output, thinking
- Available at snapshot: yes

Reasoning model with adjustable effort. Used by Codex CLI for harder bugs and longer-horizon code tasks.

- Input tokens: USD 1.1 per 1,000,000 input tokens
- Output tokens: USD 4.4 per 1,000,000 output tokens

### Veo 3.0

- Model ID: `veo-3.0`
- Model page: https://aiduct.ai/models/veo-3.0
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Google Veo 3.0 — previous generation, cheaper than Veo 3.1 for general clips.

- Video duration: USD 0.45 per 1 video second

### Veo 3.1

- Model ID: `veo-3.1`
- Model page: https://aiduct.ai/models/veo-3.1
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Latest Google video generation model with highest quality

- Video duration: USD 0.642 per 1 video second

### Veo 3.1 Lite

- Model ID: `veo-3.1-lite`
- Model page: https://aiduct.ai/models/veo-3.1-lite
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

High-efficiency Google Veo 3.1 model for lower-cost text-to-video and image-to-video generation with audio.

- Video duration: USD 0.081 per 1 video second

### Veo 3.1 Fast

- Model ID: `veo3-fast`
- Model page: https://aiduct.ai/models/veo3-fast
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Google Veo 3.1 Fast: lower-latency Veo 3.1 with native audio, first/last frames and reference images.

- Video duration: USD 0.162 per 1 video second

### Vidu Q2 Image to Video

- Model ID: `vidu-q2-image-to-video`
- Model page: https://aiduct.ai/models/vidu-q2-image-to-video
- Provider: vidu
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Vidu Q2 image-to-video route for animating a starting frame.

- Video duration: USD 0.06 per 1 video second

### Vidu Q2 Reference to Video

- Model ID: `vidu-q2-reference-to-video`
- Model page: https://aiduct.ai/models/vidu-q2-reference-to-video
- Provider: vidu
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Vidu Q2 reference image-to-video route for subject consistency with multiple reference images.

- Video duration: USD 0.06 per 1 video second

### Vidu Q3

- Model ID: `vidu-q3`
- Model page: https://aiduct.ai/models/vidu-q3
- Provider: vidu
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Q3 multi-reference image-to-video workflow for consistent subjects across scenes

- Video duration: USD 0.129 per 1 video second

### Vidu Q3 Image to Video

- Model ID: `vidu-q3-image-to-video`
- Model page: https://aiduct.ai/models/vidu-q3-image-to-video
- Provider: vidu
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Vidu Q3 image-to-video route for animating a starting frame.

- Video duration: USD 0.129 per 1 video second

### Vidu Q3 Pro

- Model ID: `vidu-q3-pro`
- Model page: https://aiduct.ai/models/vidu-q3-pro
- Provider: vidu
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Latest Vidu Q3 text and image-to-video path with audio-enabled 1-16 second clips

- Video duration: USD 0.129 per 1 video second

### Vidu Q3 Reference to Video

- Model ID: `vidu-q3-reference-to-video`
- Model page: https://aiduct.ai/models/vidu-q3-reference-to-video
- Provider: vidu
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Vidu Q3 multi-reference image-to-video route for consistent subjects across scenes.

- Video duration: USD 0.129 per 1 video second

### Wan 2.6

- Model ID: `wan-v2.6`
- Model page: https://aiduct.ai/models/wan-v2.6
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Legacy start-frame workflow for 5, 10, or 15 second clips with optional soundtrack input

- Video duration: USD 0.162 per 1 video second

### Text Embedding 3 Large

- Model ID: `text-embedding-3-large`
- Model page: https://aiduct.ai/models/text-embedding-3-large
- Provider: openai
- Capabilities: embedding
- Available at snapshot: yes

OpenAI's strongest embedding model. 3072-dim default; use when retrieval quality matters more than cost.

- Embeddings: USD 0.13 per 1,000,000 embedding tokens

### Text Embedding 3 Small

- Model ID: `text-embedding-3-small`
- Model page: https://aiduct.ai/models/text-embedding-3-small
- Provider: openai
- Capabilities: embedding
- Available at snapshot: yes

OpenAI's most cost-effective embedding model. 1536-dim default with `dimensions` truncation supported down to 256.

- Embeddings: USD 0.02 per 1,000,000 embedding tokens

### Wan 2.6 Flash

- Model ID: `wan-v2.6-flash`
- Model page: https://aiduct.ai/models/wan-v2.6-flash
- Provider: vercel
- Capabilities: video-generation, image-to-video
- Available at snapshot: yes

Faster legacy Wan start-frame workflow for lower-cost iterations

- Video duration: USD 0.129 per 1 video second

### Whisper-1

- Model ID: `whisper-1`
- Model page: https://aiduct.ai/models/whisper-1
- Provider: vercel
- Capabilities: stt
- Available at snapshot: yes

OpenAI speech-to-text transcription model exposed through the gateway /v1/audio/transcriptions primitive.

- Audio minutes: USD 0.006 per 1 audio minute

### Wan 3.0 Video

- Model ID: `wan3.0-video`
- Model page: https://aiduct.ai/models/wan3.0-video
- Provider: vercel
- Capabilities: video-generation, image-to-video, video-to-video
- Available at snapshot: yes

Alibaba Wan 3.0 all-in-one video generation with text, first/last frames, image, video, and audio references, native audio, and clips up to 30 seconds

- Video duration: USD 0.2 per 1 video second

### VibeVoice

- Model ID: `vibevoice`
- Model page: https://aiduct.ai/models/vibevoice
- Provider: fal
- Capabilities: tts
- Available at snapshot: yes

Natural text-to-speech using fal.ai's verified VibeVoice 7B presets

- Video duration: USD 0.00108 per 1 video second
- Audio seconds: USD 0.00108 per 1 audio second

### GPT-6.1 Sol

- Model ID: `gpt-6.1-sol`
- Model page: https://aiduct.ai/models/gpt-6.1-sol
- Provider: zenmux
- Capabilities: chat, vision, tools, thinking, structured-output
- Available at snapshot: yes

OpenAI's GPT-6.1 Sol for coding, computer use and professional work, with adjustable reasoning effort.

- Input tokens: USD 2 per 1,000,000 input tokens
- Output tokens: USD 10 per 1,000,000 output tokens
- Cached input: USD 0.09999999999999999 per 1,000,000 cached input tokens
- Cache writes: USD 2.5 per 1,000,000 cache write tokens

### Claude Sonnet 4.5

- Model ID: `claude-sonnet-4-5`
- Model page: https://aiduct.ai/models/claude-sonnet-4-5
- Provider: zenmux
- Capabilities: chat, vision, tools, thinking, structured-output
- Available at snapshot: yes

Anthropic's balanced flagship — strong reasoning, vision, and tool-use at a moderate price. Used by Claude Code by default.

- Input tokens: USD 3 per 1,000,000 input tokens
- Output tokens: USD 15 per 1,000,000 output tokens
- Cached input: USD 0.3 per 1,000,000 cached input tokens
- Cache writes: USD 3.75 per 1,000,000 cache write tokens
- One-hour cache writes: USD 6 per 1,000,000 one-hour cache write tokens
- Web searches: USD 0.01 per 1 web search

### Claude Opus 4.7

- Model ID: `claude-opus-4-7`
- Model page: https://aiduct.ai/models/claude-opus-4-7
- Provider: zenmux
- Capabilities: chat, vision, tools, thinking, structured-output
- Available at snapshot: yes

Anthropic's most capable model. Use for long-context analysis, ambiguous research-style problems, and the hardest agentic tasks.

- Input tokens: USD 5 per 1,000,000 input tokens
- Output tokens: USD 25 per 1,000,000 output tokens
- Cached input: USD 0.5 per 1,000,000 cached input tokens
- Cache writes: USD 6.25 per 1,000,000 cache write tokens
- One-hour cache writes: USD 10 per 1,000,000 one-hour cache write tokens
- Web searches: USD 0.01 per 1 web search

### Claude Sonnet 4.6

- Model ID: `claude-sonnet-4-6`
- Model page: https://aiduct.ai/models/claude-sonnet-4-6
- Provider: azureAnthropic
- Capabilities: chat, vision, tools, thinking, structured-output
- Available at snapshot: yes

Anthropic legacy Sonnet generation, superseded by Claude Sonnet 5; kept for workloads pinned to its behaviour.

- Input tokens: USD 3 per 1,000,000 input tokens
- Output tokens: USD 15 per 1,000,000 output tokens
- Cached input: USD 0.3 per 1,000,000 cached input tokens
- Cache writes: USD 3.75 per 1,000,000 cache write tokens
- One-hour cache writes: USD 6 per 1,000,000 one-hour cache write tokens
- Web searches: USD 0.01 per 1 web search

### Claude Opus 5.5

- Model ID: `claude-opus-5-5`
- Model page: https://aiduct.ai/models/claude-opus-5-5
- Provider: azureAnthropic
- Capabilities: chat, vision, tools, thinking, structured-output
- Available at snapshot: yes

Anthropic's most capable model, for the hardest coding, agentic, creative and long-context work.

- Input tokens: USD 4 per 1,000,000 input tokens
- Output tokens: USD 20 per 1,000,000 output tokens
- Cached input: USD 0.19999999999999998 per 1,000,000 cached input tokens
- Cache writes: USD 5 per 1,000,000 cache write tokens
- One-hour cache writes: USD 8 per 1,000,000 one-hour cache write tokens
- Web searches: USD 0.01 per 1 web search

### Claude Haiku 5.5

- Model ID: `claude-haiku-5-5`
- Model page: https://aiduct.ai/models/claude-haiku-5-5
- Provider: azureAnthropic
- Capabilities: chat, vision, tools, thinking
- Available at snapshot: yes

Fast Claude model for high-volume agent, routing and tool-use workloads. Base token pricing for prompts up to 100,000 tokens.

- Input tokens: USD 0.09999999999999999 per 1,000,000 input tokens
- Output tokens: USD 0.5 per 1,000,000 output tokens
- Cached input: USD 0.01 per 1,000,000 cached input tokens
- Cache writes: USD 0.125 per 1,000,000 cache write tokens
- One-hour cache writes: USD 0.19999999999999998 per 1,000,000 one-hour cache write tokens

### MiniMax Speech 02

- Model ID: `minimax-speech-02`
- Model page: https://aiduct.ai/models/minimax-speech-02
- Provider: minimax
- Capabilities: tts
- Available at snapshot: yes

MiniMax Speech-02 HD text-to-speech with MiniMax's multilingual system voices.

- Speech characters: USD 40 per 1,000,000 characters

### GLM-5.3 Flash

- Model ID: `glm-5.3-flash`
- Model page: https://aiduct.ai/models/glm-5.3-flash
- Provider: cloudflare
- Capabilities: chat, tools, thinking, structured-output
- Available at snapshot: yes

Z.ai's fast GLM-5.3 variant for high-volume agent turns, with reasoning and tool use at a fraction of the flagship price.

- Input tokens: USD 0.15 per 1,000,000 input tokens
- Output tokens: USD 0.5 per 1,000,000 output tokens
- Cached input: USD 0.03 per 1,000,000 cached input tokens

### GLM-5.3

- Model ID: `glm-5.3`
- Model page: https://aiduct.ai/models/glm-5.3
- Provider: zenmux
- Capabilities: chat, tools, thinking, structured-output
- Available at snapshot: yes

Z.ai's GLM-5.3 for long-context reasoning, structured output, and agent tool use.

- Input tokens: USD 1.4 per 1,000,000 input tokens
- Output tokens: USD 4.4 per 1,000,000 output tokens
- Cached input: USD 0.26 per 1,000,000 cached input tokens

### Qwen 3.8 27B

- Model ID: `qwen3.8-27b`
- Model page: https://aiduct.ai/models/qwen3.8-27b
- Provider: cloudflare
- Capabilities: chat, tools, thinking, structured-output
- Available at snapshot: yes

Alibaba's 27B instruction-tuned Qwen 3.8, tuned for agentic tool use and long-context work.

- Input tokens: USD 0.5 per 1,000,000 input tokens
- Output tokens: USD 3 per 1,000,000 output tokens
- Cached input: USD 0.049999999999999996 per 1,000,000 cached input tokens

### Claude Haiku 4.5

- Model ID: `claude-haiku-4-5`
- Model page: https://aiduct.ai/models/claude-haiku-4-5
- Provider: zenmux
- Capabilities: chat, vision, tools, structured-output
- Available at snapshot: yes

Fastest and cheapest Claude. Within ~5 percentage points of Sonnet 4.5 on many benchmarks at roughly one-third the cost.

- Input tokens: USD 1 per 1,000,000 input tokens
- Output tokens: USD 5 per 1,000,000 output tokens
- Cached input: USD 0.09999999999999999 per 1,000,000 cached input tokens
- Cache writes: USD 1.25 per 1,000,000 cache write tokens
- One-hour cache writes: USD 2 per 1,000,000 one-hour cache write tokens
- Web searches: USD 0.01 per 1 web search

### DeepSeek Chat

- Model ID: `deepseek-chat`
- Model page: https://aiduct.ai/models/deepseek-chat
- Provider: openrouter
- Capabilities: chat, tools, structured-output
- Available at snapshot: yes

Strong reasoning at a very low price point. Competitive with GPT-4o-mini for code and structured-output workloads.

- Input tokens: USD 0.14 per 1,000,000 input tokens
- Output tokens: USD 0.28 per 1,000,000 output tokens
- Cached input: USD 0.014 per 1,000,000 cached input tokens

### Flux Dev

- Model ID: `flux-dev`
- Model page: https://aiduct.ai/models/flux-dev
- Provider: fal
- Capabilities: image-generation
- Available at snapshot: yes

Development model with good quality-speed balance

- Images: USD 0.042 per 1 image

### Flux Schnell

- Model ID: `flux-schnell`
- Model page: https://aiduct.ai/models/flux-schnell
- Provider: vercel
- Capabilities: image-generation
- Available at snapshot: yes

Fastest Flux model for rapid iteration

- Images: USD 0.018 per 1 image

### Claude Sonnet 5

- Model ID: `claude-sonnet-5`
- Model page: https://aiduct.ai/models/claude-sonnet-5
- Provider: azureAnthropic
- Capabilities: chat, vision, tools, thinking, structured-output
- Available at snapshot: yes

Anthropic's most capable Sonnet-class model, built for coding, agents, professional work, and long-context multimodal tasks.

- Input tokens: USD 2 per 1,000,000 input tokens
- Output tokens: USD 10 per 1,000,000 output tokens
- Cached input: USD 0.19999999999999998 per 1,000,000 cached input tokens
- Cache writes: USD 2.5 per 1,000,000 cache write tokens
- One-hour cache writes: USD 4 per 1,000,000 one-hour cache write tokens
- Web searches: USD 0.01 per 1 web search

### Gemini Audio

- Model ID: `gemini-2.5-flash-preview-tts`
- Model page: https://aiduct.ai/models/gemini-2.5-flash-preview-tts
- Provider: gemini
- Capabilities: tts
- Available at snapshot: yes

Gemini native single-speaker text-to-speech with Gemini's prebuilt voices.

- Requests: USD 0.018 per 1 request

### GLM-5.2

- Model ID: `glm-5.2`
- Model page: https://aiduct.ai/models/glm-5.2
- Provider: zenmux
- Capabilities: chat
- Available at snapshot: yes

GLM-5.2 routed by the Aiduct API gateway.

- Input tokens: USD 0.42 per 1,000,000 input tokens
- Output tokens: USD 1.32 per 1,000,000 output tokens
- Cached input: USD 0.078 per 1,000,000 cached input tokens

### MiniMax Music V2

- Model ID: `minimax-music-v2`
- Model page: https://aiduct.ai/models/minimax-music-v2
- Provider: minimax
- Capabilities: music-generation
- Available at snapshot: yes

High-quality AI music and sound effect generation. Async-only — POST returns a polling ticket like /v1/videos/generations because generation takes 30s-2min upstream.

- Requests: USD 0.048 per 1 request

### Recraft V3

- Model ID: `recraft-v3`
- Model page: https://aiduct.ai/models/recraft-v3
- Provider: vercel
- Capabilities: image-generation
- Available at snapshot: yes

Professional design and illustration generation

- Images: USD 0.066 per 1 image

### Seedream 3.0

- Model ID: `seedream-v3`
- Model page: https://aiduct.ai/models/seedream-v3
- Provider: volcengine
- Capabilities: image-generation
- Available at snapshot: yes

ByteDance Seedream 3.0 text-to-image model for fast creative drafts and general image generation.

- Images: USD 0.03 per 1 image

### SD 3.5 Large

- Model ID: `stable-diffusion-v35-large`
- Model page: https://aiduct.ai/models/stable-diffusion-v35-large
- Provider: fal
- Capabilities: image-generation
- Available at snapshot: yes

Stable Diffusion 3.5 Large model

- Images: USD 0.105 per 1 image

### TTS-1

- Model ID: `tts-1`
- Model page: https://aiduct.ai/models/tts-1
- Provider: vercel
- Capabilities: tts
- Available at snapshot: yes

OpenAI's standard text-to-speech model. Fast, natural-sounding voices (alloy, echo, fable, onyx, nova, shimmer). Use for conversational latency-sensitive workloads.

- Speech characters: USD 15 per 1,000,000 characters

### TTS-1 HD

- Model ID: `tts-1-hd`
- Model page: https://aiduct.ai/models/tts-1-hd
- Provider: vercel
- Capabilities: tts
- Available at snapshot: yes

Higher-fidelity OpenAI text-to-speech. Same voices as tts-1 with cleaner audio; pick this for narration and consumer-facing audio.

- Speech characters: USD 30 per 1,000,000 characters

### Vidu Q2 Text to Video

- Model ID: `vidu-q2-text-to-video`
- Model page: https://aiduct.ai/models/vidu-q2-text-to-video
- Provider: vidu
- Capabilities: video-generation
- Available at snapshot: yes

Vidu Q2 text-to-video route with fast 2-8 second generation.

- Video duration: USD 0.06 per 1 video second

### Vidu Q3 Text to Video

- Model ID: `vidu-q3-text-to-video`
- Model page: https://aiduct.ai/models/vidu-q3-text-to-video
- Provider: vidu
- Capabilities: video-generation
- Available at snapshot: yes

Vidu Q3 text-to-video route for prompt-only generation.

- Video duration: USD 0.129 per 1 video second