Utopai Studios' Utopai X Climbs to No. 2 in Both Video Leaderboards

Utopai Studios has extended Utopai X into image-to-video generation. In the latest cited snapshot of the Artificial Analysis leaderboard , the model ranks second behind Dreamina Seedance 2.5. It also holds second place in text-to-video, making it the only model ranked in the top two on both boards.

Utopai X starts with MiniMax H3 and receives additional post-training, meaning Utopai adapts an existing model rather than training one from scratch. The studio says it developed and evaluated the model against production requirements and feedback from filmmakers and artists. Its ranking offers evidence that targeted post-training can push a base video model beyond the original release and competing variants.

Two boards, one contender

AA-Video-I2V v2.1 measures how well models animate a source image into video with audio. Evaluators compare pairs of outputs, and their preferences determine each model’s Elo rating. Elo measures relative performance within the tested field, while scores and positions can change as new votes arrive.

MiniMax H3 at 768p leads the benchmark’s open-weights models with an Elo score of 1,126. Utopai X ranks above that base model, showing the measurable effect of Utopai’s additional training and evaluation.

Strengths favor motion and structure

Artificial Analysis divides results into capabilities, which test specific model skills, and use cases, which group outputs by their intended application. Utopai X leads two image-to-video capability categories and remains close to the leaders in three others.

Use-case results follow the same pattern. Utopai X leads UI/UX & Motion Design, including kinetic typography, interface demonstrations, and abstract motion graphics. It also leads Productivity & Knowledge Work, which includes explainers, educational videos, animated data visualizations, and scenario simulations. The model narrowly leads the Frontier and Retail & Ecommerce categories and sits just behind Dreamina Seedance 2.5 in Marketing & Advertising.

Human-centered video remains the gap

Dialogue & Lip Sync and Human Anatomy produce Utopai X’s largest deficits against the leading models. Live-Action Film, covering cinematic shots, microdrama episodes, and previsualization, is another weaker category. These results make the model less suitable for dialogue-heavy scenes and character-driven footage that depends on precise facial movement or anatomy.

Utopai X closes the gap to the category leader in seven of ten capabilities when compared with MiniMax H3. Its largest gains appear in Physics, Human Anatomy, and Audio Synchronization. Human Anatomy has improved substantially over the base model while still trailing the strongest competitors.

PAI access sets the limit

Utopai offers the model exclusively through PAI, its production platform, with no public API. Developers therefore cannot call Utopai X directly from automated generation pipelines or integrate it through a standard model endpoint.

A ten-second 1080p clip consumes 930 credits, leaving little capacity on the entry plan for retries or additional generations. Teams must also adopt PAI’s workflow rather than connecting the model to their existing infrastructure.

Post-training drives the result

Utopai X demonstrates how production-focused post-training can alter the competitive position of an established base model. The same approach is common in language models, where teams adapt a general model with specialized data, preference feedback, and evaluation criteria. Video developers are now applying that pattern to motion, narrative continuity, audio timing, and production-specific output.

The model’s position above MiniMax H3 and other H3-based variants makes the post-training result more relevant than the ranking alone. Teams can improve a capable foundation model for specific creative workloads without absorbing the cost of pretraining a new video model.

Choose by workload

• Consider Utopai X for motion graphics, interface demonstrations, explainers, animated data, multi-scene narratives, and physics-heavy sequences when PAI fits the production workflow.
• Compare category leaders for dialogue, lip sync, human anatomy, and character-driven live-action work.
• Use an API-accessible model for automated pipelines. The base MiniMax API provides a more direct integration path.