Chinese AI models win image and video share in Overchat AI data

New Overchat AI usage data shows Chinese models taking the lead in image and video generation, while U.S. providers still dominate text. The shift matters because it suggests users choose different model makers depending on the task, device and market.

Why it matters: - Overchat AI’s data suggests model choice is splitting by use case, with Chinese labs gaining ground in visual generation while U.S. providers retain strength in text. - The pattern shows that access to multiple models in one app can change user preferences over time, especially as people try alternatives. - The findings also hint that pricing and availability may matter less than performance and familiarity in some categories.

What happened: - Overchat AI analyzed anonymized activity from August 24 to October 5, 2026, covering 1.09 million users, 1.2 million sessions and 31.6 million interactions. - Chinese AI models accounted for 36.9% of all AI interactions on Overchat AI. - Chinese models reached 43.2% of usage in the U.S., 44.6% in the U.K., 46.4% in Australia and 52.0% in Spain. - In image generation, Chinese models held 59.5% of usage, up from 54.0% in the prior period. - In video generation, Chinese models rose to 53.6% from 43.5%. - Text remained dominated by U.S. providers, with Gemini, GPT, Claude and Grok combining for 86.2% of text-model interactions.

The details: - OpenAI and Anthropic made up 23.8% of new users’ model usage, compared with 19.8% among returning users. - Google’s share rose from 20.9% among new users to 28.0% among returning users. - Chinese providers increased from 36.3% of new users’ usage to 39.8% among returning users. - Google and Chinese models combined for 67.8% of returning users’ usage. - ByteDance’s Seedream led image-model families with 33.5%. - Google’s Nano Banana followed at 26.1%. - Alibaba’s Qwen Image reached 26.0%, more than doubling from 12.3% in the earlier period. - From September 22 to October 5, Qwen Image became the most-used model family across all categories at 17.2%, ahead of GPT text models at 10.9% and Gemini at 10.8%. - OpenAI’s GPT Image reached 7.8% of image-model usage after GPT Image 2.5 launched on September 10. - Qwen Image and GPT Image 2.5 cost the same on Overchat AI and were both included in the free plan. - Qwen Image was used about four times as often as GPT Image during the final two weeks of the period. - In video, Grok Imagine remained the largest individual family at 30.6%, down from 39.6%. - ByteDance’s Seedance climbed to 25.2% from 10.8%, driven by Seedance 2.5. - Kuaishou’s Kling held 21.9% of video usage. - Seedance 2.5 cost roughly three times as much per video as Grok Imagine 1.5. - In Japan, Gemini generated 74.4% of AI usage. - Spain and Indonesia favored Qwen Image. - The U.S. was the only major market where video was the leading category, at 36.8%. - Brazil was the most text-focused market, at 49.3%. - Mobile accounted for about 74% of sessions. - On mobile, usage was split nearly evenly across text at 33.3%, image at 33.6% and video at 33.1%. - Desktop usage was more text-heavy at 51.3%. - Desktop’s leading families were GPT, Gemini and Claude. - Mobile favorites were Seedream, Grok Imagine and Seedance. - By developer, Google led with 23.1% of all model usage. - ByteDance followed with 17.0%, OpenAI with 13.6% and xAI with 13.0%. - OpenAI posted the largest increase, from 8.8% to 13.6%, supported by GPT-6 and GPT Image 2.5 releases. - Overchat AI said usage shares count each user once per model used during the period, rather than counting every interaction. - Overchat AI said model families combine available versions and provider origin reflects the developer’s location. - Overchat AI said no individual accounts, prompts, uploads or generated content were reviewed.

Between the lines: - The split between text and visual tasks suggests users may trust different companies for different creative jobs. - The stronger showing for Chinese models in image and video implies competition in generative AI is not uniform across product types. - The rise in returning-user share for Google and Chinese providers suggests model switching increases after initial trial. - The data also shows that free access and matching price do not guarantee similar adoption.

What's next: - Overchat AI is likely to keep tracking how releases from Google, OpenAI, ByteDance and others shift usage shares. - Future updates will show whether Qwen Image’s surge and Seedance’s video gains hold as newer models roll out. - The company’s next comparisons will help show whether regional and device-based preferences stay stable or continue to move.