Skip to content

Models

Official prices, key limits, and available groups for every public model. Open any model for its full spec.

Claude

10

Native Anthropic-protocol models that work directly with Claude Code and the Messages API.

GPT Image

3

Billed per token; the only image family with mask-based partial edits, and it can output transparent backgrounds.

GPT

10

Native Codex and OpenAI SDK models, available on both Responses and Chat Completions.

Gemini

1

The only model that returns video synchronously: one request returns an MP4, no polling.

Gemini Image

5

Generates by aspect ratio, takes up to 14 reference images, billed per delivered image.

Gemini

13

OpenAI Chat Completions compatible; one API key covers the whole Flash, Lite, and Pro lineup.

GLM

2

1M context with thinking always on; available on both Chat Completions and Responses.

Qwen

5

All five models have a 1M context window and 131,072 max output, and one group covers them all.

DeepSeek

3

1M-context reasoning models; thinking can be turned on or off, and prefix completion is supported.

Grok Imagine

1

Per-second text-to-video with optional reference images, 1–15 seconds long.

Grok Imagine

3

Priced per image, two tiers (1k / 2k), up to 2 reference images for edits.

Grok

5

Context up to 2M with reasoning always on; available on both Chat Completions and Responses.

Seedream

3

Priced per image, synchronous only, 1 image per request; well suited to large 2K–4K images.

Seedance

14

Text-to-video and multimodal-reference video; the overseas 2.0 standard models are the only ones that support 4K.

Kling Image

7

Async image generation with up to 10 reference images, plus dedicated models for image-to-image, multi-image reference, and outpainting.

Kling

13

Billed per second; covers first and last frames, subject reference, motion control, avatars, and lip sync.

Midjourney

1

Always 4 images per task; aspect ratio and style parameters go in the prompt.

MiniMax Speech

2

Synchronous text to speech, billed per character, with official voices and voice cloning.

MiniMax Hailuo

2

24 FPS video with native stereo audio; dialogue and narration are generated with the picture.

Wan · HappyHorse

4

Wan and HappyHorse: native task API, billed per second, up to 30 seconds long.

Prices are official prices in USD. Actual charge = official price × group multiplier; see Usage in the console for exact charges.