MiniMax H3 Pricing, Cost & API Guide

MiniMax H3 Pricing: Output and Input Materials

Input and output rates for MiniMax H3 differ by task: video output is billed by the second, while H3-Context-IR text tokens are billed by the million. On the pay-as-you-go API, video output costs $0.08 per second at 768P and $0.13 per second at 2K. Total costs also depend on audio, image, and reference-video inputs. There are 6 billing line items to understand before estimating costs:

The H3-Context-IR figures are the only H3-related text-token figures in the official rate card. Context-IR analyzes supplied multimodal context and returns an enhanced prompt. It is not the cost of the MiniMax H3 video-generation model.

Video Cost Examples

A 5-second H3 video at 768P costs $0.40; the same clip at 2K costs $0.65. The table below applies the official per-second output prices and excludes billable extra image inputs or reference-video input.

MiniMax H3 cost scales linearly with output duration. Choosing 768P over 2K cuts the per-second output charge by roughly 38%. Starting prompt testing at 768P and short durations, then rendering selected candidates at 2K, is the most direct lever for controlling spend during development.

Supported Generation Modes and Limits

MiniMax H3 supports 5 named working modes across 3 broad workflows: text-to-video, first/last-frame image-to-video, and reference generation. These modes are T2VA, I2VA, FL2VA, L2VA, and Ref2VA.

MiniMax H3 outputs 768P or 2K video, with duration set as an integer between 4 and 15 seconds. Prompts may run up to 7,000 characters. Text-to-video requires a chosen ratio of 21:9, 16:9, 4:3, 1:1, 3:4, or 9:16. Image-to-video derives its ratio from the input image, while reference generation can use adaptive or an explicit supported ratio.

Reference-generation limits include up to 9 images, 3 video clips, 3 audio clips, and 12 mixed files total. Reference videos and audio clips must each be 2–15 seconds. Total reference-video duration and total reference-audio duration are each capped at 15 seconds.

How to Get MiniMax H3 API Access

MiniMax H3 API access starts at the MiniMax developer platform, where account creation, key generation, and balance funding all live in one place. There are 4 sequential steps to reach a working API call:

1. Create a MiniMax Account

MiniMax H3 access begins with registering at the MiniMax developer platform using a valid email address. Account verification is required before the console dashboard becomes accessible.

2. Fund the Pay-as-You-Go Balance

MiniMax H3 requires a funded pay-as-you-go account. Add credits to the account balance before making generation requests. Unfunded accounts return a 402 error on every call.

3. Generate an API Key

MiniMax H3 API keys are generated from the API Keys section inside the MiniMax console. Select Create new key, assign a label for tracking purposes, and copy the key immediately — the console displays the full key string only once.

4. Note the Base URL and Endpoint

The MiniMax H3 API base URL is https://api.minimax.io. The create-task endpoint is POST /v2/video_generation. Every request requires the API key passed as a Bearer token in the Authorization header.

MiniMax H3 API keys should be stored in an environment variable rather than hardcoded into source files, to prevent accidental credential exposure in version control.

MiniMax H3 API Quickstart: Submit, Poll, and Retrieve Video

The MiniMax H3 API accepts a POST request to /v2/video_generation, with a JSON body containing the model name, a content array, resolution, duration, and, for text-to-video, ratio. The minimal working call requires 2 headers and 4 body fields.

Authentication and Required Headers

Every MiniMax H3 request requires an Authorization header formatted as Bearer <YOUR_API_KEY>. The second required header is Content-Type: application/json.

Python Quickstart: Submit, Poll, and Retrieve Video

MiniMax H3 requests are asynchronous. A create request returns a task_id. Poll the task-status endpoint and continue until the task succeeds, fails, or is cancelled:

import os

import time

import requests

API_KEY = os.environ["MINIMAX_API_KEY"]

BASE_URL = "https://api.minimax.io"

headers = {

    "Authorization": f"Bearer {API_KEY}",

    "Content-Type": "application/json",

}

payload = {

    "model": "MiniMax-H3",

    "content": [

        {

            "type": "text",

            "text": "A paper airplane glides through a sunlit library; slow forward camera movement."

        }

    ],

    "resolution": "2K",

    "duration": 5,

    "ratio": "16:9",

}

create = requests.post(f"{BASE_URL}/v2/video_generation", headers=headers, json=payload)

create.raise_for_status()

task_id = create.json()["task_id"]

while True:

    status = requests.get(

        f"{BASE_URL}/v2/query/video_generation/{task_id}", headers=headers

    )

    status.raise_for_status()

    task = status.json()["task"]

    if task["status"] == "succeeded":

        print(task["content"]["url"])

        break

    if task["status"] in {"failed", "cancelled"}:

        raise RuntimeError(task.get("error", "Video generation did not succeed"))

    time.sleep(10)

In this request, model identifies the H3 checkpoint. The content array carries the text prompt. The resolution, duration, and ratio fields control output format. On success, task.content.url contains the generated video URL.

How to Add Images, Last Frames, and References

Role-tagged content items enable image-to-video requests in MiniMax H3. For a first-frame request, add an image_url content item with role: "first_frame". For a first-and-last-frame request, add one item with role: "first_frame" and another with role: "last_frame".

For a reference-generation request, use reference_image, reference_video, or reference_audio roles. First/last-frame and reference workflows cannot be mixed within the same request.

A concise first/last-frame motion description should state what connects the anchors. For example: "The camera holds a wide shot as the subject walks from the window to the desk. Daylight fades to evening without a cut." The official H3 skill advises describing the relationship between keyframes while keeping timing consistent with the requested 4–15-second duration.

Cost-Control Checklist

There are 4 cost-control practices that reduce MiniMax H3 spend without sacrificing output quality:

• Start at 768P. Test prompts at 768P before committing to 2K renders. The per-second rate is 38% lower.
• Keep image inputs within the free tier. The first 5 images per request are free. Stay within that limit where the project permits.
• Use reference video selectively. Reference-video input is billed by input-video duration at the output resolution rate. Use it only when it provides enough creative value to justify the charge.
• Log every request. Record output duration, resolution, image count, and reference-video seconds on each call. This lets you reconstruct API cost from the published rate card without relying on dashboard estimates.

Common MiniMax H3 API Errors and Troubleshooting

The MiniMax H3 API returns a 401 error when the API key is missing or invalid — fix it by setting Authorization: Bearer YOUR_API_KEY on every request.

There are 6 error categories developers encounter most often with the MiniMax H3 API:

The official error reference documents a 429 rate-limit response but does not promise a Retry-After header. Applications should implement bounded retry and logging rather than assume a particular response header.

For a browser-based way to test the model before building an API workflow, open Minimax H3.

Frequently Asked Questions

Is MiniMax H3 an LLM priced by input and output tokens?

No. MiniMax H3 is a video model with 768P and 2K video output priced per generated second. H3-Context-IR is a separate prompt-enhancement task that has token pricing; it does not generate video.

What does MiniMax H3 cost for a 10-second video?

At published pay-as-you-go list prices, a 10-second output costs $0.80 at 768P or $1.30 at 2K before applicable input-material charges.

Can MiniMax H3 use first and last frames?

Yes. The video API supports a prompt plus a first-frame image, last-frame image, or both.

Does MiniMax H3 support reference audio?

Yes. Reference generation accepts reference images, video, and audio. The H3 prompt-writing skill includes audio-focused guidance.

How long can a MiniMax H3 output be?

The documented duration range is 4–15 seconds in integer values.

How do I get an API key for MiniMax H3?

MiniMax H3's API key is generated inside the MiniMax developer console after account registration. The key must be passed in the Authorization header of every request.

What payment methods does MiniMax offer for H3?

MiniMax H3 billing uses a pay-as-you-go prepaid balance. Users deposit credits into their MiniMax account; each API call deducts from that balance in real time. Enterprise and volume pricing options are available by contacting MiniMax directly.

Why am I getting a 401 or 429 error when calling the MiniMax H3 API?

MiniMax H3's API returns 2 common errors. A 401 error signals an invalid or missing API key in the Authorization header. A 429 error signals that the account has exceeded its rate limit. Confirm the key is correctly formatted, then implement bounded retry with back-off before retrying.