Skip to content

    MiniMax-H3

    Video generation

    MiniMax-H3

    MiniMax-H3 on FlexAI: MiniMax Video Gen, MiniMax H3 Community License, $4.8 / min of video on the OpenAI-compatible Token Factory.

    Pricing and availability

    MiniMax-H3 rates on FlexAI, in US dollars. Billed per usage unit with no minimum.
    RatePrice
    Price$4.8 / min of video
    AvailabilityServerless, on the shared Token Factory API

    Use of MiniMax-H3 is subject to the MiniMax H3 Community License, including its Acceptable Use Policy. Content you publish with it must be disclosed as AI-generated.

    Deployment requirements

    API endpoint
    /v1/videos/generations
    Compatibility
    OpenAI
    Parameters
    33.1B
    Active parameters
    33B of 33.122993B
    Quantization (served)
    BF16
    Native precision
    BF16, from safetensors
    Weights
    MiniMaxAI/MiniMax-H3
    Published by
    MiniMax
    Origin
    China

    Estimate your monthly cost

    minutes
    1,000 × $4.8/min of video$4,800
    Metered per second$0.08/s
    Estimated monthly cost$4,800

    Estimate only, at the current market-tracked rate. Usage-based; no minimums.

    Get an API key

    How to call MiniMax-H3

    # NOTE: video generation is asynchronous and is not covered by
    # the OpenAI SDK. Use raw httpx (or requests) to POST then poll.
    import base64
    import os
    import time
    import httpx
    
    HEADERS = {"Authorization": f"Bearer {os.environ['FLEXAI_API_KEY']}"}
    
    # 1. Submit the job. POST returns 202 + a job id.
    resp = httpx.post(
        "https://api.flex.ai/v1/videos/generations",
        headers={**HEADERS, "Content-Type": "application/json"},
        json={"model": "MiniMax-H3", "prompt": "a cat walking through a forest"},
        timeout=30,
    )
    resp.raise_for_status()
    job_id = resp.json()["id"]
    print(f"Submitted {job_id}, polling…")
    
    # 2. Poll until completed. A clip usually takes one to a few minutes.
    #    The completed response carries the clip, and it is served once.
    while True:
        r = httpx.get(f"https://api.flex.ai/v1/videos/generations/{job_id}", headers=HEADERS, timeout=30)
        r.raise_for_status()  # 410 if the clip was already collected or expired
        j = r.json()
        if j["status"] == "completed":
            break
        if j["status"] == "failed":
            raise RuntimeError(f"Generation failed: {j.get('error')}")
        time.sleep(10)
    
    # 3. Decode the base64 mp4 and save it.
    video_bytes = base64.b64decode(j["data"][0]["b64_json"])
    with open("out.mp4", "wb") as f:
        f.write(video_bytes)
    print("Saved out.mp4")

    Swap models, not your agent

    MiniMax-H3 runs behind the same OpenAI-compatible key as every other model in the catalog. Change the model string and your skills, tools, scoped memory, and governance carry over unchanged. The same setup scales to a dedicated endpoint or your own private cloud.