Claude Fable 5.2 now costs less. Improvements to our serving architecture have lowered the cost of using the model, and the new rates are live.See new pricing
Minimax

MiniMax H3

Save 50%Official

minimax/minimax-h3

MiniMax H3 — MiniMax's new open multimodal video model: 4–15 sec, 768P or 2K, text-to-video, first/last frame, plus reference images, videos and audio.

Live model playground

Try MiniMax H3 here

Tune the request, see the current price ceiling, and inspect the real response without leaving this model page.

asyncvideo
1Configure input
59 / 20,000
First frame
Last frame
Reference images

Sign in first to upload images to this vision model.

Maximum estimated cost—
2Live resultExample
Example output · generated Oct 24, 2026 · took 175.7s · cost $0.2
Input modalities
TextImage
Output modalities
Video
Capabilities
Video generation · Text-to-video · Image-to-video · Reference voices · Reference images · Reference videos · First and last frame
Upstream release
2026-07-31
RouteMux name
minimax/minimax-h3
Official ID
MiniMax-H3

What is MiniMax H3

MiniMax H3 is MiniMax's next-generation open, general-purpose multimodal video model, released on July 31, 2026. It reads a text prompt together with images and turns them into a video clip. In its launch post MiniMax says it set aside the Hailuo-02 architecture for H3 and built a single generation model that understands text, image, video and audio context together.

Three ways to start: from a prompt alone (text-to-video), from a first frame or a first and last frame (image-to-video), or from references: up to nine reference images, three reference videos and three reference audio clips (videos and audio each 2–15 seconds per clip, 15 seconds in total) that set the character, motion, camera, style or voice. Frame images and references cannot be mixed in one request. Duration is any whole number of seconds from 4 to 15, output is 768P or 2K, and the aspect ratio can be 16:9, 9:16, 1:1, 4:3, 3:4 or 21:9; with an input image or reference video it follows the input by default. Prompts can run to 7,000 characters, and camera instructions such as [pan], [zoom] or [static] can follow a key description.

Compared with Hailuo 2.3, length is no longer limited to 6 or 10 seconds, resolution moves from 768P / 1080P to 768P / 2K, and last frames, reference images, reference videos and reference audio are new inputs. If you need 1080P output, or a pipeline is built around Hailuo 2.3's fixed tiers, Hailuo 2.3 remains available.

Pricing is per output second, with one rate for 768P and another for 2K, so a 10-second clip costs ten times the per-second rate. Reference video adds a charge per input second at the same rate as the chosen output tier; the first five reference images are free and each one after that is charged; reference audio is free. It uses the OpenAI Sora protocol and is asynchronous: the request returns a job, and you poll it for the finished video.

MiniMax H3 API pricing

OutputOfficial $0.08 /sec
$0.04/secSave 50%
  • 768P$0.04 /secOfficial $0.08
  • 2K$0.065 /secOfficial $0.13
  • Reference video · 768P$0.04 /secOfficial $0.08
  • Reference video · 2K$0.065 /secOfficial $0.13
  • Reference images (first 5 free, then each)$0.02Official $0.04

Key capabilities

4–15 seconds
Any whole-second length, not fixed presets.
768P or 2K
Two output resolutions, each with its own per-second rate.
Frames or references
First frame, first + last frame, or up to 9 reference images, 3 reference videos and 3 audio clips.
Asynchronous
Returns a job; poll it to get the video.
50% below official pricing
Billed per output second; both per-second rates are in the pricing section above.

Limited time10% extra Credits on your first top-up — up to 200 bonus Credits

Applies to your first top-up within 72 hours of signing up. Minimum $20, once per account.

Sign up and claimNo subscription. Pay for what you use.

Interface and protocols

Use any of these IDs to call this model via the API.

RouteMux nameminimax/minimax-h3
Official IDMiniMax-H3

Replace the ROUTEMUX_KEY placeholder with your API key. Create one →

curl https://api.routemux.com/v1/videos \
  -H "Authorization: Bearer <ROUTEMUX_KEY>" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax/minimax-h3",
    "prompt": "A quiet product shot of a glass observatory at dusk",
    "seconds": 5,
    "size": "1280x720"
  }'
base_url: https://api.routemux.com/v1

Performance (last 30 days)

MiniMax H3 FAQ

›What is MiniMax H3?

MiniMax H3 — MiniMax's new open multimodal video model: 4–15 sec, 768P or 2K, text-to-video, first/last frame, plus reference images, videos and audio.

›How much does the MiniMax H3 API cost?

Through RouteMux, MiniMax H3 is billed per second, by resolution: 768P $0.04, 2K $0.065. Each is below the official list price.

›How do I call MiniMax H3 via API?

MiniMax H3 is OpenAI-compatible: point your base URL at RouteMux and set the model field to minimax/minimax-h3 — no code changes needed.

›Which input and output modalities does MiniMax H3 support?

MiniMax H3 accepts text, image as input and produces video as output.

›When was MiniMax H3 released?

MiniMax H3 was released on July 31, 2026.

Sources