How the ranking is constructed
The ranking cited by MarkTechPost is based on a single quality indicator: the Elo derived from blind votes cast by certified evaluators in Artificial Analysis's arena, where each vote compares two clips generated from the same prompt. The fixed reference is Kling 3.0 1080p Pro at 1,000 points on both tables, text-to-video and image-to-video: a lower score means that voters more often preferred the Kuaishou model.
Alongside this figure, the authors include the price per second derived from per-minute rates normalized to 1080p and the weight-license status. A methodological caveat accompanies the results: most scores in the top 10 have a confidence interval of between ±13 and ±15 Elo points, so models within about 25 points should be considered statistically close and positions three through eight as a group, not a rigid order.
The leaderboard overview
Alibaba's Wan 3.0 leads the text-to-video ranking with 1,175 points, backed by 7,651 votes, and was launched on August 24, 2026: it generates 30-second clips at 1080p, even starting from documents and presentations, at $0.20 per second, but its weights are not downloadable. Utopai X, developed by Utopai Studios by further training MiniMax H3, is the only model in the top 2 of both tables, with 1,170 points in text-to-video and 1,151 in image-to-video; it is accessible only through its creator's PAI platform, at 93 credits per second, and cannot be invoked from a pipeline.
MiniMax H3's third place marks the entry of an open-weight model into the leading group: 1,161 points in text-to-video and 1,133 in image-to-video, at a price of $0.08 per second and with a license described as open but with restrictions. The model, a dense 33-billion-parameter Transformer with a Qwen3-VL-32B encoder and native 32 kHz stereo audio, has become, according to the source, the basis for numerous derivatives: Utopai X, HeyGen Video 1, fal's H3 Max, and Pika H3 Flex.
Next are ByteDance's Dreamina Seedance 2.5, which leads the image-to-video table with 1,161 points and supports up to 30 seconds with multiple references, and Google's Gemini Omni Flash 1.2, available from November 5, 2026 with clips of 3 to 15 seconds and a 30 percent price cut compared with the previous version. The upper tier also includes FLUX 3 Max, HeyGen Video 1, H3 Max, SkyReels V4, and FLUX 3.
What changes for buyers
The most relevant finding for businesses concerns the quality-to-cost ratio: according to the source, the three cheapest leading models are H3 or its derivatives, and quality and price no longer move in tandem. The article calculates an indicative value by combining the Elo advantage over the 1,000 threshold and the price per second, specifying that the comparison ignores differences in clip resolution and duration.
Some points remain to be verified: rates vary with resolution, and some license terms for the open weights are not detailed in the available material. API availability in some geographic regions and the release timing of the open-weight versions announced by Black Forest Labs for FLUX 3 are described as not yet defined and must be confirmed before setting an adoption strategy.
