← All models

MiniMax H3

Published Publisher MiniMaxMiniMax-H3

Overview

MiniMax H3 is a general-purpose multimodal generation model focused on video with synchronized native stereo audio. It understands combined text, image, video and audio context and supports instruction-following, reference-based generation and editing. MiniMax announced it on July 31, 2026, and released model weights on August 3, 2026. (minimax.io)

Capabilities and uses

  • Generates 4–15-second videos at 24 fps, with 32 kHz stereo audio. The open H3-Base produces 768p short-side output; H3-Regenerate-2K can regenerate it at 2K.
  • Supports text-to-video, first- and last-frame control, and multimodal references. Advertised applications include advertising, branding, e-commerce, product design, UI/UX and gaming.
  • The reference variant accepts up to 9 images, 3 video clips and 3 audio clips, with at most 12 files in total. Each video or audio clip must be 2–15 seconds; each modality has a 15-second total-duration limit. Audio references cannot be used without an image or video reference.
  • Stable dialogue support is stated for 11 languages: Arabic, Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian and Spanish. (minimax.io)

Limitations and access

The released weights use the MiniMax H3 Community License; they are not an unconditional grant of commercial rights. MiniMax states that commercial products or services exceeding USD 20 million in annual revenue require separate written authorization, and commercial interfaces must prominently display the model name. The hosted H3-Context-IR preprocessing system is not included in the open-weight release. MiniMax also identifies multimodal understanding and visual detail as areas for improvement. (design.minimax.io)

Official API pricing

Prices below are official platform list prices checked November 16, 2026. Domestic CNY and international USD prices are separate regional offerings, not currency conversions. Video generation is billed by duration and input materials, rather than by input/output tokens. (platform.minimax.cn)

Mainland China platform — CNY

  • MiniMax-H3 generation: 2K output CNY 0.80 per output second and input video CNY 0.80 per input second; 768p output CNY 0.50 per output second and input video CNY 0.50 per input second. The input-video rate depends on output resolution.
  • Input materials: audio is free; the first 5 images per task are free, then CNY 0.20 per additional image.
  • Optional 768p-to-2K regeneration: CNY 0.30 per regenerated output second, plus CNY 0.30 per second of video input from the original task; the first 5 original input images are free, then CNY 0.15 per additional image. Original input audio is free. This is charged separately from the original task.
  • H3-Context-IR: CNY 5.80 per million input tokens and CNY 23.00 per million output tokens.
  • Official domestic pricing. (platform.minimax.cn)

International platform — USD

  • MiniMax-H3 generation: 2K output USD 0.13 per output second and input video USD 0.13 per input second; 768p output USD 0.08 per output second and input video USD 0.08 per input second. The input-video rate depends on output resolution.
  • Input materials: audio is free; the first 5 images per task are free, then USD 0.04 per additional image.
  • Optional 768p-to-2K regeneration: USD 0.05 per regenerated output second, plus USD 0.05 per second of video input from the original task; the first 5 original input images are free, then USD 0.025 per additional image. Original input audio is free. This is charged separately from the original task.
  • H3-Context-IR: USD 0.90 per million input tokens and USD 3.60 per million output tokens.
  • Official international pricing. (platform.minimax.io)

Official sources