Back to Devoured - August 04, 2026
DEVOURED
China's MiniMax H3 is the first open model to top an AI video ranking

China's MiniMax H3 is the first open model to top an AI video ranking

AI llmopensource The Decoder
MiniMax's H3 is the first open-weights model to top Artificial Analysis's video leaderboards, though it leaves some core modules closed-source.
What: The 33-billion-parameter H3 model supports text, image, video, and audio inputs for four- to 15-second generation. It is licensed for commercial use only to companies with under $20 million in revenue and lacks the H3-Context-IR and 2K resolution modules found in the proprietary version.
Why it matters: This indicates that while open-weights models are reaching parity with closed systems in performance, vendors are increasingly relying on 'partial openness'—releasing core inference logic while keeping pre-processing pipelines or high-resolution upscalers proprietary.
Deep dive
  • 33-billion-parameter architecture supports multi-modal inputs including reference images and audio clips.
  • Ranked first in video editing, second in text-to-video, and third in image-to-video benchmarks.
  • Weights are available on Hugging Face, enabling fine-tuning for custom styles or characters.
  • ComfyUI integration allows local execution, currently limited to 768p resolution.
  • Commercially restrictive license targets smaller entities under $20M revenue.
  • ByteDance launched the competing closed-source Seedance 2.5 on the same day.
Decoder
  • Context-IR: An intermediate representation layer that parses prompts and reference media into a structured format for the generative model.
  • ComfyUI: A node-based GUI for Stable Diffusion and other generative models, allowing users to build complex inference workflows.
Original article

MiniMax releases H3 video model weights, putting an open model at the top of a video ranking for the first time. Artificial Analysis ranks H3 first in Video Editing, second in Text-to-Video, and third in Image-to-Video. The 33-billion-parameter model processes text, images, video, and audio together, generating four- to 15-second clips with stereo sound. According to the model card, a single prompt can include up to nine reference images, three video clips, and three audio clips.

Two pieces remain closed, though. The 2K resolution module and H3-Context-IR, which translates prompts and reference material into a structured intermediate format, aren't included. Running H3 locally in ComfyUI tops out at 768p, and users will need to handle context prep themselves using MiniMax's published prompting guides. The open weights do allow fine-tuning on custom footage, characters, or a specific visual style. One catch on the license side: commercial use is only permitted for companies making under $20 million in revenue.

ByteDance released its closed Seedance 2.5 the same day, which generates 30-second clips with built-in audio.

Explore all 45 articles in today's digest

Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi