MiniMax H3模型权重已公开，支持全模态理解与原生立体声音频

TL;DR — MiniMax H3 (the official name of what most people call Hailuo 3.0) is the third generation of MiniMax's Hailuo video line, but it deliberately stops behaving like a video model. It reads text, images, video, and audio as one unified context, then generates a 4–15 s clip at 2K/24 fps with native stereo audio — no separate audio stage, no post-hoc upscaler. It supports up to 9 reference images, 3 reference video clips, and 3 reference audio clips per generation. On the Artificial Analysis leaderboards it ranks #1 in Video Editing, #2 in Text-to-Video, #3 in Image-to-Video. Weights are promised "in the coming days" under a planned MiniMax Community License (commercial use for organizations under $20M revenue, with attribution), but no Hugging Face model card exists yet as of August [...] MiniMax H3 is the official model name; Hailuo 3.0 — also written Hailuo 03 — is the widely used alias, after the Hailuo AI app it ships in. They are one model, the direct successor to Hailuo 2.3.

To reduce the computational cost of long multimodal sequences, H3 natively supports sparse-attention training and inference. The initial open-source release provides inference with full attention only. Our sparse-attention implementation will be released in a future update. [...] The model uses three-dimensional Multimodal Rotary Position Embeddings (MM-RoPE) to represent positional relationships across the temporal and two spatial dimensions, `(t, h, w)`. During the final stage of training, we introduce native sparse attention to reduce the computational cost of long sequences. The sparse-attention implementation is not included in the initial open-source release and will be published separately in a future update. [...] # MiniMax H3 ## System Overview MiniMax H3 is a general-purpose, omni-modal generative system. It supports unified understanding of multimodal contexts composed of text, images, video, and audio, and can generate video with native stereo audio at resolutions up to

MiniMax MiniMax M3MiniMax M2.7MiniMax M2.5 视频生成视频生成 MiniMax H3 语音&音乐语音&音乐 MiniMax Speech 2.8MiniMax Music 3.0 产品 MiniMax CodeMiniMax Hub语音星野 API Token Plan 研究 关于我们 与所有人共创智能与所有人共创智能 公司介绍新闻投资者关系加入我们 EN MiniMax 模型 语言模型语言模型 MiniMax M3NEWMiniMax M2.7MiniMax M2.5 视频生成视频生成 MiniMax H3NEW 语音&音乐语音&音乐 MiniMax Speech 2.8NEWMiniMax Music 3.0NEW 产品 MiniMax CodeNEWMiniMax Hub语音星野 API Token Plan 研究 关于我们 与所有人共创智能与所有人共创智能 公司介绍新闻投资者关系加入我们 English # MiniMax H3：打破任务和模态的边界 AIVideo GenerationMultimodalH3 今天，我们正式发布 MiniMax H3。这是一款通用的全模态生成模型，支持对文本、图像、视频、声音组成的多模态上下文的统一理解能力、能够输出具备原生双声道的音视频，最高可支持 15s 2K 分辨率。 根据前期的邀测反馈，MiniMax H3 具备商用级的多场景内容生成能力，在指令遵循、文字与品牌信息呈现、V2V Motion Transfer（视频到视频动作迁移）等方面表现出色，可实现精准、可控的多模态内容编辑与生成，广泛适用于广告、品牌、电商、产品设计、UI/UX、游戏等商业场景。

Sign uphere. MiniMax said ⁠it planned to release H3's model weights within days, allowing users to ​download and customise the underlying system. The move would extend the open-weight ​approach increasingly adopted by Chinese AI developers into video generation, where many prominent models remain proprietary. Founded in 2022, MiniMax is one of China's "AI tigers", a group of well-funded ​startups competing with domestic tech giants and U.S. leaders in the crowded ​market for foundation models. It became the second of the group to list publicly ‌when ⁠it floated in Hong Kong in January, following Z.AI(2513.HK), opens new tab.

MiniMax completed its Hong Kong IPO in January 2026, raising approximately $619 million at an approximately $6.5 billion IPO valuation, with Alibaba and Tencent among its backers. H3 follows its M3 large language model (released June 2026) and continues the company's pattern of pairing technical announcements with open-weight commitments — commitments that, in the case of M3, shipped weights approximately 17 days after the API launch. H3 is currently available via the MiniMax API (model ID: MiniMax-H3) and the Hailuo AI consumer platform. Model weights are expected within days, the company said — subject to applicable laws and regulations. Read more: Seedance 2.5 API Is Live: ByteDance's 30-Second AI Video Carries Unresolved Copyright Risk ## Frequently Asked Questions [...] MiniMax released H3 today — a multimodal video generation model that ranks first in video editing among all models tracked by independent benchmarking firm Artificial Analysis — and simultaneously announced plan

Log inSign up ## Post user avatar MiniMax (official) @MiniMax\_AI MiniMax-H3 Is Now Publicly Available huggingface.co/MiniMaxAI/Mini… 00:00 2:39 AM · Aug 3, 20261.5MViews user avatar MiniMax (official) @MiniMax\_AI 11h Online API Use MiniMax-H3 directly via API. Global: platform.minimax.io | CN: platform.minimaxi.com Online App Use MiniMax-H3 directly via App. WebApp Global: hailuoai.video | CN: hailuoai.com Desktop Global: hub.minimax.io | CN: Models - MiniMax API DocsFrom platform.minimax.io 4.4K user avatar RyanLee MiniMax (official) @RyanLeeMiniMax Aug 3 [...] MiniMax (official) @RyanLeeMiniMax Aug 3 About Licence: This regional carve-out stems from our ongoing generative video copyright litigation with major @Hollywood studios. We aim to deliver our AI tools globally to all users responsibly. US-based persons looking to deploy MiniMax-H3 may submit formal licensing 353K