AB
AiBoss
Tutorials

PPIO releases "Top Ten Trends in Domestic Large Model Call Volume in the First Half of 2025" (PDF file)

The report, "Top Ten Trends in the Call Volume of Domestic Large-Scale Models in the First Half of 2025," summarizes the top ten trends in the call volume of domestic large-scale models in the first half of 2025. The report points out that DeepSeek and Qwen have become the two most popular open-source models in China, particularly in non-inference models...

Domestic products in the first half of 2025Large ModelThe "Top Ten Trends in Call Volume" summarizes the domestic call volume trends in the first half of 2025.Large ModelTop 10 Trends in Call Volume. The report indicates that DeepSeek and Qwen have become the two most popular apps in China.open sourceThe use of non-inference models is gradually surpassing that of inference models. In the fields of image and video generation, user preferences differ, with significant differences in the usage ratios of text-to-image and image-to-video models. Stable Diffusion XL and Alibaba Cloud's Wanxiang have become the most popular image and video generation model infrastructures, respectively. The report mentions a surge in demand for overseas expansion, with approximately 20% of...AIApplication scenarios are geared towards overseas markets.intelligentbodyApplications are also availablefastOutbreak. According to the PPIO platform...Large Model The token usage volume summarizes the domestic situation in the first half of 2025.Large ModelTop 10 trends.

Get PPIODomestic production in the first half of 2025Large ModelTop 10 Trends in Call VolumeOriginal PDF file, scan the QR code to follow and reply: 20250717

DeepSeek and Qwen each occupy half of the market share, making them the two most popular search engines in China.open sourceModel

DeepSeek and Qwen are the two most popular apps in China.open sourceIn terms of models, DeepSeek once accounted for 99% of the usage share on the PPIO platform, but with the emergence of more models, Qwen's usage has grown rapidly since the second quarter of 2025, forming a duopoly with DeepSeek.

Although inference models, exemplified by DeepSeek R1, have led the technological paradigm of reinforcement learning, non-inference models like DeepSeek V3 have become more popular.

Although inference models such as DeepSeek R1 have led the technological paradigm, non-inference models such as DeepSeek V3 have surpassed inference models in usage since March 2025 due to their wide applicability and low barrier to entry, and have long accounted for more than 60% of the market.

DeepSeek's mathematical and distillation models have a relatively small user base; its greater significance lies in driving industry research.

DeepSeek's mathematical models, Prover V2 and Distillation Model, are used relatively little, accounting for only 1% of the DeepSeek family, but they are crucial for its advancement.AIBasic research in the industry is of great significance.

freeThe number of calls to Qwen 2.5 is significantly higher than that of the more powerful Qwen 3, indicating that price remains one of the most important factors for users when choosing a software model, compared to model performance.

On the PPIO platform,freeQwen 2.5 has a much higher call volume than the more powerful Qwen 3, and its token consumption accounts for more than 90% of the total Qwen series, indicating that price is one of the important factors for users to choose.

In image generation models, the ratio of text-generated images to image-generated images is approximately 9:1.

Among image generation models, the text-based image generation model saw its usage share increase from 70% in January to 90% in June due to its low creation threshold and high-quality generated images, far exceeding that of the text-based image generation model.

In the video generation model, the ratio of text-based videos to image-based videos is approximately 1:9.

Among video generation models, users prefer the image-generated video model, with its usage increasing from about 20% in January to over 90% in June, because image-generated videos are highly controllable, stable, and in line with creators' habits.

Stable Diffusion XL is gradually replacing Stable Diffusion 1.5, becoming the most popular image generation model infrastructure.

Stable Diffusion XL has seen rapid growth in usage since late March due to its large number of parameters, excellent generation results, and strong community support. It has gradually replaced Stable Diffusion 1.5 and become the most popular image generation model infrastructure.

AliWanxiang is the most popular in Chinaopen sourceVideo Model

Since its launch in March 2025, Alibaba's Wanxiang platform has experienced rapid growth in usage, accounting for over 80% of the PPIO platform's traffic by June, becoming the most popular domestic video platform.open sourceThe MiniMax-Hailuo-02 video model also achieved 1.5% usage in its first week of release.

Demand for overseas expansion is surging, with approximately 20% of users...AIApplication scenarios are geared towards overseas markets.

worldwideAIThe market has great potential, and overseas users are interested in it.AIGC products have a strong willingness to pay, with approximately 20% of users on the PPIO platform paying for them.AIApplication scenarios are geared towards overseas markets and domestic markets.AIThe team and developers are actively expanding overseas.

2025,intelligentbodyBig Outbreak

In the first half of 2025,intelligentbodyApplications such as Manus, Flowith, etc.fastbreak out,Large ModelIt has already penetrated various industries, and the PPIO platform will also be released soon.Agent Sandbox products helpintelligentbodydevelop.

Get PPIODomestic production in the first half of 2025Large ModelTop 10 Trends in Call VolumeOriginal PDF file, scan the QR code to follow and reply: 20250717

Kimi+ said,AIOne-clickGenerate professional 3D digital human micro-courses

Alibaba Cloud ChinaartificialintelligentApplication Development Report (2025) (PDF file)