MiniMax H3: Complete Specifications, Pricing, API Access & Use Cases (2026)

MiniMax H3 is a general-purpose multimodal generation model from MiniMax, designed to combine text, images, video, and audio in creative video workflows rather than treating each media task as a separate model. MiniMax launched H3 publicly on July 31, 2026 and subsequently released its model weights. This guide focuses on the specifications, pricing, production implications, Gate.AI access, and comparisons that matter to developers and creative teams evaluating H3.

What Is MiniMax H3?

MiniMax H3 is an open-weight video generation model capable of interpreting multimodal context and producing video with native stereo audio. MiniMax positions it as a broader successor to task-specific Hailuo workflows: instead of separating text-to-video, image-to-video, reference, motion-transfer, audio, and editing tasks completely, H3 uses natural-language instructions to describe relationships between several input modalities.

The provider documents text, image, video, and audio context, video output up to 15 seconds, resolutions up to 2K, text and brand rendering, instruction-guided editing, and video-to-video motion transfer. MiniMax open-sourced H3 on August 3, 2026, including FL2VA and Ref2VA checkpoints for local inference.

Gate.AI separately lists H3 with the catalog identifier minimax/minimax-h3. Its documented video API quick start uses minimax-h3 as the request-body model value, an important distinction when integrating the API.

What Are MiniMax H3’s Key Specifications and Pricing?

Gate.AI shows a 131,072 context value, but MiniMax does not describe this as a conventional LLM-style maximum prompt window in its H3 launch material. It is therefore safer to treat 131K as Gate.AI listing metadata rather than a provider-confirmed text-token limit.

Pricing is easier to translate into a production estimate. At the listed 768p rate, a 10-second video costs approximately 10 × $0.08 = $0.80. If one reference image is billed, the listed image-input charge raises the calculation to approximately $0.84 before any additional task-specific charges. A 10-second 2K output would be approximately $1.30, excluding applicable input charges.

What Can MiniMax H3 Do That Makes It Useful in Production?

H3 is particularly relevant when a creative brief contains several kinds of references. MiniMax demonstrates workflows where motion comes from a reference video, a character comes from an image, and audio supplies another part of the target scene. Natural-language instructions describe how those sources should interact.

This can reduce the need to route separate generation, reference-transfer, and audiovisual tasks through unrelated models. Advertising and e-commerce teams can also use H3’s documented text and brand-rendering focus for product clips, animated posters, interface concepts, opening titles, and branded creative.

A second production advantage is deployment flexibility. MiniMax has released H3 model weights and documents support paths involving SGLang, vLLM, diffusers, and ComfyUI, while Gate.AI provides a managed asynchronous API route. Teams can therefore evaluate hosted generation against self-managed inference rather than being restricted to one access model.

What Are MiniMax H3’s Supported Modalities?

MiniMax describes H3 as an omni-modal system that jointly understands text, image, video, and audio context. The provider documents 4–15 second output, multiple aspect ratios, and resolution up to 2K. Gate.AI’s generic video endpoint additionally supports reference-image, reference-video, and reference-audio fields, although developers should verify the exact combination accepted for their H3 workflow before production deployment.

Where Does MiniMax H3 Fall Short?

H3 remains a short-form video model: MiniMax’s documented maximum generation length is 15 seconds, so long-form production still requires multiple generations and downstream editing.

Several operational fields also remain less explicit than they would be for a mature text-model API. Public H3 materials do not establish the Gate.AI 131K value as a conventional text prompt window, and public benchmark claims should not be treated as equivalent to independent production testing.

The Gate.AI model card also presents both minimax/minimax-h3 and the API request value minimax-h3. Developers should follow the model-specific quick-start schema rather than assuming the catalog path itself is the executable request ID.

What Is MiniMax H3 Best Used For?

MiniMax H3 fits workflows where multiple creative references must be combined into one short audiovisual output. Examples include branded advertising with visible text, product demonstrations driven by a reference image, motion transfer from an existing clip, animated UI or website concepts, and audiovisual social content.

Choose H3 when multimodal references, native audio, editing flexibility, 2K output, or access to open weights are important. For simpler short text-to-video or image-to-video generation where those broader reference capabilities are unnecessary, a narrower model may provide a more straightforward workflow.

Human review remains appropriate for brand spelling, product details, visual continuity, audio accuracy, copyright-sensitive source material, and commercially important claims.

How Does MiniMax H3 Compare to Hailuo 2.3 and Wan 2.6 T2V?

H3 overlaps with existing video-generation pages, but its search intent should remain distinct: it is a general-purpose multimodal reference and editing model, whereas Hailuo 2.3 primarily targets short text/image-to-video generation and Wan 2.6 T2V is centered on text-to-video creation.

Readers focused on MiniMax’s earlier short-video workflow can compare the Hailuo 2.3 specifications, pricing, and API guide. For a narrower competing text-to-video workflow, see the Wan 2.6 T2V specifications and pricing guide.

How Do I Access MiniMax H3 Through Gate.AI?

Gate.AI documents H3 through its asynchronous video-generation endpoint:

POST https://api.gate.ai/api/v1/videos

Authentication uses a Gate.AI API key in the Authorization: Bearer header. The H3 model page’s executable quick start uses minimax-h3 as the request model value. Jobs return a job_id, which can then be queried through GET /api/v1/videos/{job_id}.

Python Example

import os
import requests
api_key = os.environ["GATEAI_API_KEY"]
headers = {
    "Authorization": f"Bearer {api_key}",
    "Content-Type": "application/json",
}
payload = {
    "model": "minimax-h3",
    "prompt": "A premium product ad, slow camera orbit, clean studio lighting",
    "duration": 6,
    "resolution": "720p",
    "aspect_ratio": "16:9",
    "generate_audio": True
}
response = requests.post(
    "https://api.gate.ai/api/v1/videos",
    headers=headers,
    json=payload,
    timeout=60
)
response.raise_for_status()
data = response.json()["data"]
print("Job ID:", data["job_id"])
print("Status URL:", data["status_url"])

curl Example

curl "https://api.gate.ai/api/v1/videos" \
  -H "Authorization: Bearer $GATEAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax-h3",
    "prompt": "A premium product ad, slow camera orbit, clean studio lighting",
    "duration": 6,
    "resolution": "720p",
    "aspect_ratio": "16:9",
    "generate_audio": true
  }'

These examples follow Gate.AI’s documented H3 quick start and video task schema; they are documentation-based examples, not execution-tested examples.

FAQs

Is MiniMax H3 open source?

MiniMax describes H3 as open source and released model weights on August 3, 2026. The published checkpoints include FL2VA and Ref2VA task families and can be used through supported inference frameworks.

Does MiniMax H3 generate audio?

Yes. MiniMax documents native stereo audio generated alongside video rather than requiring a separate audio-generation step.

What is MiniMax H3’s maximum video length?

MiniMax’s open-weight documentation lists generated video durations from 4 to 15 seconds.

How much does MiniMax H3 cost on Gate.AI?

Gate.AI lists 768p video generation at $0.08 per second, 2K generation at $0.13 per second, and image input at $0.04 per image as of September 2026.

Does MiniMax H3 have a 131K context window?

Gate.AI displays a context value of 131,072. MiniMax’s provider documentation, however, does not define H3 using a conventional LLM token-context specification, so the value should not automatically be interpreted as a 131K text-prompt limit.