MiniMax Hailuo H3 API: Official Price per Second, Model IDs and How to Call It

Short answer

TeamoRouter supports MiniMax H3, the video model MiniMax runs in its Hailuo AI app. You call it with the OpenAI-compatible /v1/videos API and one of three model IDs: minimax-h3-text-to-video, minimax-h3-image-to-video or minimax-h3-reference-to-video.

• What it makes: video with native stereo sound, up to 15 seconds at 2K (MiniMax's H3 post). 2K is the default. 768P is the cheaper option.
• Inputs: a text prompt, a first frame (with an optional last frame), or reference images.
• MiniMax's official API price: $0.13 per second of output at 2K and $0.08 at 768P (MiniMax pricing). A 10-second 2K clip is about $1.30.
• On TeamoRouter: no MiniMax account needed. You pay per job from a balance that starts at $5. The same key also calls text and image models.

New accounts start with a $0 balance — top up before your first request. Top up from $5. If a request returns 402 (400 on the OpenAI format) with "wallet balance is insufficient", your balance is empty — your base URL and key are fine. Top up and retry.

Get a key and generate video with MiniMax H3

What MiniMax H3 does

MiniMax released H3 on July 31, 2026. From its H3 post and API reference:

• One context for every input. H3 reads text, images, video and audio together. You say in plain words how the inputs relate to the clip you want, for example "use the camera move from video 1 and have the person in image 2 sing".
• Sound comes with the picture. Audio is generated together with the video, and all of it is native stereo.
• 2K by default. For 2K, H3 regenerates its own lower-resolution output instead of using a separate upscaler. MiniMax says this keeps small text and fine detail.
• Reference and editing in one model. Text-to-video, image-to-video with first and last frames, and reference-to-video all run on H3. You describe the edit or the reference in the prompt.
• The next Hailuo generation. MiniMax describes H3 as the generation after Hailuo 01 and Hailuo 02.

Call MiniMax H3 through TeamoRouter

Video jobs are asynchronous: create a job, poll it, then download the MP4. Python with the OpenAI SDK:

import os, time
from openai import OpenAI

client = OpenAI(api_key=os.environ["TEAMO_API_KEY"], base_url="https://api.teamorouter.cn/v1")
video = client.videos.create(
    model="minimax-h3-text-to-video",
    prompt="A night market food stall, the cook tosses noodles in a flaming wok, sizzling sound and crowd noise",
    seconds="10",
    size="2560x1440",  # 2K, 16:9. Use "1365x768" for 768P.
)
while video.status in ("queued", "in_progress"):
    time.sleep(5)
    video = client.videos.retrieve(video.id)
if video.status == "failed":
    raise RuntimeError(video.error)
client.videos.download_content(video.id).write_to_file("h3.mp4")

Image-to-video with curl. Pass the first frame in input_reference:

curl https://api.teamorouter.cn/v1/videos \
  -H "Authorization: Bearer $TEAMO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax-h3-image-to-video",
    "prompt": "The woman in the photo turns to the camera and laughs, wind moves her hair, quiet street sound",
    "seconds": "6",
    "input_reference": {"image_url": "https://example.com/first-frame.png"}
  }'

• The response has a job id. Poll GET /v1/videos/{id} until status is completed, then download GET /v1/videos/{id}/content. Finished videos are kept for seven days.
• Always send seconds as a string: "4" to "15" for H3, "5" to "15" for the H3 Max IDs below.
• size sets resolution and shape: 2560x1440 is 2K at 16:9, 1365x768 is 768P at 16:9, and 1440x2560 is 2K at 9:16. Without size, you get 768P. For image-to-video, MiniMax takes the frame shape from your image.
• Optional JSON fields: end_image_url adds a last frame to image-to-video. audio is on by default; send "audio": false for a silent clip. For reference-to-video, input_reference can be a list of images.
• GET /v1/models?type=video lists the current model IDs and the durations and sizes each one accepts. Full reference: video generation API.
• Prefer a UI? Use the video workbench with the same balance.

Which model ID to use

• Need 2K: use the three standard H3 IDs. The H3 Max IDs stop at 768P on TeamoRouter.
• Testing prompts: start at 768P with short clips, then render the final take at 2K.
• H3 Max: TeamoRouter lists it as a separate model with its own price. See the H3 Max model page.

How much MiniMax H3 costs

MiniMax's pricing page bills H3 per second of output video:

• The price is the same for text-to-video, image-to-video and reference-to-video.
• The first 5 input images per job are free. Each extra image costs $0.04.
• Input video is billed by its length at the same per-second rate. Audio input is free.

TeamoRouter's current price for each H3 model ID is on the MiniMax H3 model page and the pricing page.

Paying for usage

No subscription: you top up a balance from $5 by card. Each video job is charged to that balance. The same key works for every model in the catalog.

FAQ

Is MiniMax H3 the model in Hailuo AI?

Yes. The Hailuo AI site says MiniMax H3 is live, and MiniMax's H3 post places it after Hailuo 01 and Hailuo 02. Through TeamoRouter you call it by model ID and pay per job, with no Hailuo membership.

How much does the MiniMax H3 API cost?

MiniMax charges $0.13 per second at 2K and $0.08 per second at 768P. A 15-second 2K clip is about $1.95. TeamoRouter's price is on the model page.

How long and how sharp can an H3 video be?

4 to 15 seconds per job, at 768P or 2K. On TeamoRouter, the H3 Max IDs take 5 to 15 seconds at 768P.

Is MiniMax H3 open source?

MiniMax has published H3 weights on Hugging Face under its own community license. Per MiniMax's self-hosting guide, the open release is H3-Base, which makes 768p video. The 2K step is not included, and the license excludes some regions. If you want 2K without running GPUs yourself, use the API.

Related

• MiniMax H3 model page
• MiniMax M3 API: price and how to call it
• Video generation API
• All video models