MiniMax Launches Open-Source Video Model H3: Direct 2K Output, Tops Editing Arena
量子位 · wechat · 2026-07-31
MiniMax has officially released its next-generation video model, MiniMax H3, marking it as their first open-source video model. Breaking the traditional paradigm where models only generate raw clips, H3 integrates editing logic, typography, transitions, background music, and VFX end-to-end, supporting full-modal input to directly output 2K resolution, publish-ready videos.
In hands-on testing, H3 demonstrated strong intent comprehension and multimodal synergy. It effortlessly generates hand-drawn effects and dynamic titles while significantly improving text rendering in videos—a known AI weak point—ensuring stability and lighting consistency across frames. It also supports audio input for emotion-matched dialogue generation.
Pricing is highly competitive, with 2K generation costs at less than 1/3 of mainstream models per second. Technically, the team built the H3-OmniTransformer architecture, utilizing a custom Caption model to achieve deep multimodal emergence. Its open-source nature provides enterprises with a top-tier model for private deployment and custom fine-tuning.
Related event: MiniMax Releases Multimodal Model H3 with 2K Native Stereo Video(18 posts)→
More from Models
- Reflection's 501B MoE Beam hits 463 tok/s on 8x GB200 with NVFP4 at day 0 — zhyncs42 · 2026-10-16
- JevBench v1.7.46 leaderboard: new d1-ultra tops 32 API systems at 76.1 composite — airesearch12 · 2026-10-16
- LMArena launches Agent Arena and Code Arena leaderboards for model comparison — arena · 2026-10-16
- Third-party tests find NVIDIA Nemotron-3.5 Bolt underperforms nano on pandas/numpy and decision-making tasks — norenEnmotalen · 2026-10-16
- Reflection open-sources Beam under Apache 2.0 with 40+ partner ecosystem — bhutanisanyam1 · 2026-10-16
- One Prompt, 629 Lines: IQuest-Q1 Builds a Deep-Sea Submarine Game in 2 Minutes — Div_pradeep · 2026-10-16