Chinese text-to-video models sweep SuperCLUE rankings as Anthropic touts “Claude Mythos”

Results: domestic models take the lead

China’s homegrown text-to-video models have surged past overseas rivals in a freshly published SuperCLUE benchmark, it has been reported that the April 2026 automated evaluation put domestic systems in the top three slots. PixVerse V6 from Aishi Technology (爱诗科技) scored 76.00 to take first place, with ByteDance (字节跳动)’s model and Shengshu Technology (生数科技)’s Vidu Q3 Pro close behind, forming a clear first tier. The automated test covered 13 mainstream domestic and international models and scored them on six dimensions including motion fluidity, content consistency, physical realism, anime, realism and fantasy, aiming for a more objective comparison.

Specific strengths differentiated the leaders. ByteDance’s model topped content consistency and sequential coherence, PixVerse V6 excelled in physical realism and stylized outputs (anime and fantasy), and MiniMax’s Hailuo 2.3 led on photorealistic generation. Notably, Google’s Veo 3.1 — the best-performing overseas model in the set — placed fifth, already showing a gap to the top domestic contenders. The report emphasizes that Chinese models have improved both base capabilities and stylized generation; motion smoothness and frame stability are now at parity with, or better than, many foreign offerings.

Significance and wider context

Why does this matter beyond model bragging rights? Because the contest is unfolding amid broader tech competition and trade frictions. As Western export controls and geopolitical tensions constrain access to some advanced components and models, domestic advances reduce reliance on foreign stacks and shift leverage toward local vendors. It has been reported that Anthropic — the US AI lab behind Claude — recently announced a new “Claude Mythos” model and claimed extraordinary code- and hacking-related capabilities; reportedly the model will not be opened to the public. Such announcements underscore a bifurcated global market where both sides race on capability and access.

The SuperCLUE results underscore a fast-maturing Chinese model ecosystem that now competes head-to-head with leading Western labs on important generative tasks. For global producers and enterprise buyers the question is practical: which models meet performance, cost and regulatory needs? For policymakers the question is strategic: how will standards, export policy and domestic investment shape the next phase of multimodal AI? The SuperCLUE ranking is one more indicator that those debates are only going to intensify.