Utopai Studios launches Utopai X video model on MiniMax H3

On 2026-10-29, MarkTechPost reported that MiniMax had open-sourced the MCode Stack repository under the MIT license. The repository collects the SKILL.md skills and checklists that the MiniMax Code team uses in daily work, covering code review, subagent splits, context budgets and long-running commands, and it is meant to be reused directly by developers.

On 2026-10-30, the benchmarking outfit Artificial Analysis disclosed on its official X account that Utopai Studios had released Utopai X, a video generation model post-trained on MiniMax H3. The model is available inside PAI, Utopai's production platform, and now supports Image to Video. The same day's evaluation put Utopai X at number two on the AA-Video-I2V v2.1 leaderboard, behind only Dreamina Seedance 2.5. Each metric came from prompts written to test a specific model skill: Artificial Analysis said Utopai X leads on Physics and Multi-Scene & Narrative in Image to Video, and sits close behind the leaders in Audio Synchronization, Spatio-Temporal Consistency and Text Rendering. Measured against real work tasks, the model leads on UI/UX & Motion Design and on Productivity & Knowledge Work, and tracks the frontier models closely in Retail & Ecommerce. Artificial Analysis invited readers to the AA-Video-I2V v2.1 and AA-Video-T2V v2.1 video leaderboards for the detailed data on image-to-video and text-to-video performance.

Also on 2026-10-30, OSChina reported on the same MiniMax Code open-source release, saying the team had open-sourced 16 internal Agent skills and workflows covering high-standard code maintenance, deciding whether a task should be handed to a subagent, and long-session optimization. The two reports disagree on the count: MarkTechPost said the repository carries 17 SKILL.md skills and 5 checklists, while OSChina said 16 Agent skills and workflows were released. The event currently rests at the published evaluations and leaderboard placements.