SuperAPI AI Directory

PPIO "Top 10 Trends in Domestic Large Model API Calls in the First Half of 2025" (PDF File) - AI Tutorial Resources

Author: AI小集 · Original source: AI工具集

"Top 10 Trends in Domestic Large Model API Calls in the First Half of 2025" summarizes the top 10 trends in domestic large model API calls 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, and the usage of non-reasoning models is gradually surpassing that of reasoning models. In the fields of image and video generation, user preferences differ, and the proportions of API calls for text-to-image models and image-to-video models differ significantly. Stable Diffusion XL and Alibaba Wanx have respectively become the most popular image and video generation model architectures. The report mentions that demand for going global is high, with about 20% of AI application scenarios oriented toward overseas markets, and agent applications are also rapidly exploding. Based on the large model tokens API call volume on the PPIO platform, the top 10 trends in domestic large models in the first half of 2025 are summarized.

Top 10 Trends in Domestic Large Model API Calls in the First Half of 2025 CakeGrowth QR Code

Get the PPIO "Top 10 Trends in Domestic Large Model API Calls in the First Half of 2025" PDF original file, scan the QR code to follow and reply: 20250717

Trend 1

DeepSeek and Qwen each occupy half the market and are the two most popular open-source models in China

DeepSeek and Qwen are the two most popular open-source models in China. DeepSeek once accounted for 99% of usage on the PPIO platform, but as more models emerged, Qwen's usage grew rapidly starting in the second quarter of 2025, forming a duopoly with DeepSeek.

Top 10 Trends in Domestic Large Model API Calls in the First Half of 2025

Trend 2

Although reasoning models represented by DeepSeek R1 have led the technical paradigm of reinforcement learning, non-reasoning models such as DeepSeek V3 are more popular

Although reasoning models such as DeepSeek R1 lead the technical paradigm, non-reasoning models such as DeepSeek V3 saw usage surpass reasoning models starting in March 2025 due to their broad applicability and low usage threshold, accounting for more than 60% for a long period.

Top 10 Trends in Domestic Large Model API Calls in the First Half of 2025

Trend 3

DeepSeek's math model and distilled models have relatively low user usage; their greater significance lies in advancing industry research

DeepSeek's math model Prover V2 and distilled models have relatively low usage, accounting for only 1% of the DeepSeek series, but they are of great significance for advancing basic research in the AI industry.

Top 10 Trends in Domestic Large Model API Calls in the First Half of 2025

Trend 4

The free Qwen 2.5 has far higher API call volume than the more powerful Qwen 3, indicating that compared with model performance, price is still one of the factors users care about most when selecting models

On the PPIO platform, the free Qwen 2.5 has far higher API call volume than the more powerful Qwen 3, with its token consumption accounting for more than 90% of the Qwen series overall, indicating that price is one of the important factors in user model selection.

Top 10 Trends in Domestic Large Model API Calls in the First Half of 2025

Trend 5

Among image generation models, the API call volume ratio of text-to-image to image-to-image is about 9:1

Among image generation models, text-to-image models, due to their low creation threshold and high generated image quality, increased their usage share from 70% in January to 90% in June, far higher than image-to-image models.

Top 10 Trends in Domestic Large Model API Calls in the First Half of 2025

Trend 6

Among video generation models, the API call volume ratio of text-to-video to image-to-video is about 1:9

Among video generation models, users prefer image-to-video models, whose usage share increased from about 20% in January to more than 90% in June, because image-to-video offers high controllability and good stability, in line with creators' habits.

Top 10 Trends in Domestic Large Model API Calls in the First Half of 2025

Trend 7

Stable Diffusion XL is gradually replacing Stable Diffusion 1.5 to become the most popular image generation model architecture

Due to its large parameter count, superior generation quality, and strong community support, Stable Diffusion XL has seen rapid growth in usage since late March, gradually replacing Stable Diffusion 1.5 to become the most popular image generation model architecture.

Top 10 Trends in Domestic Large Model API Calls in the First Half of 2025

Trend 8

Alibaba Wanxiang is the most popular open-source video model in China

Since its launch in March 2025, Alibaba Wanxiang has experienced rapid usage growth, reaching over 80% on the PPIO platform by June, becoming the most popular domestic open-source video model, while the MiniMax-Hailuo-02 video model also gained 1.5% usage in its first week online.

Top 10 Trends in Domestic Large Model API Calls in the First Half of 2025

Trend 9

Overseas expansion demand is high, with about 20% of users' AI application scenarios oriented toward overseas markets

The global AI market has great potential, and overseas users have a strong willingness to pay for AIGC products. About 20% of AI application scenarios on the PPIO platform are oriented toward overseas markets, as domestic AI teams and developers actively expand abroad.

Top 10 Trends in Domestic Large Model API Calls in the First Half of 2025

Trend 10

2025, the great explosion of agents

In the first half of 2025, agent applications such as Manus and Flowith rapidly exploded. Large models have penetrated into all industries, and the PPIO platform will also launch the Agent Sandbox product to support agent development.

Top 10 Trends in Domestic Large Model API Calls in the First Half of 2025 CakeGrowth QR code

To obtain the PDF original file of PPIO's "Top 10 Trends in Domestic Large Model API Calls in the First Half of 2025", scan the QR code to follow and reply: 20250717