The Zhitong Finance App learned that CMB International released a research report saying that after reviewing the development of the 3Q26 Internet model, judging the upper limit of the model's ability is still the key to determining differentiated pricing capabilities. Reduced cost per task and agentization are expected to support rapid growth in call volume. High value, attractive ROI, verifiable output results, sufficient data, and controllable error costs are key factors driving the further improvement of AI penetration in vertical scenarios, and the deepening of RSI is expected to further increase the depth of AI use.
CMB International's main views are as follows:
Reviewing the development of the 3Q26 industry
1) The intelligence capabilities of the basic model continue to improve. Claude Opus 5.5 and Sonnet 5.5 lead the world. Chinese manufacturers maintain their advantages in the field of open source models, but competition in open source models is also intensifying; 2) industry exploration breakthroughs are aimed at strengthening computer operation, long-range software engineering and network security capabilities to promote the expansion of programming and agents to a complete workflow; 3) increasing the scale of users and increasing the depth of use are jointly promoting commercialization progress, while major domestic model makers are expected to further broaden their revenue sources through the sharing of calls with overseas cloud platforms. Looking ahead,
Industry progress: computer operation, long-range software engineering and network security are the main breakthroughs
The 3Q26 model's ability to perform real tasks has been further improved: 1) Computer operation and result editing are gradually integrated, and products such as Cowork connect data reading, cross-software operation and document delivery to further expand agents from programming to general office; 2) software engineering capabilities extend from code generation to long-term autonomous execution, and the model can continuously develop, test, and fix problems around the same goal; 3) network security applications expand to vulnerability detection and automatic repair. CMB International believes that improving the ability to complete long-range tasks will expand the scope of work that AI can undertake and provide support for more frequent and continuous calls.
Industry trends: The number of users is expanding, and the depth of use of leading paying groups is increasing
In terms of user size, ChatGPT's weekly activity increased from 1 billion in August 2026 to 1.2 billion in October. The percentage of US consumers in the sample that paid to subscribe to at least one of ChatGPT, Gemini, or Claude also rose from 2.1% in August 2025 to 4.5% in August 2026. Using the depth dimension, B-side high-intensity users continued to increase their investment. In the Ramp US enterprise sample, the monthly AI expenditure per employee in August 2026 reached 7,205 US dollars, +198% YoY, and the top 10% and median enterprise expenses increased 246% and 165%, respectively. In terms of C-side payment penetration, the spending of leading payers is growing faster. YipitData data shows that from January 2025 to August 2026, the average monthly expenses of the top 1% AI payers in the US increased by 79%. Construction and automation, content creation, and transaction processing are important payment scenarios.
Business model: The commercialization of overseas models continues to be deepened, and domestic model monetization channels are further broadened
The main marginal changes in commercialization of 3Q26 overseas AI platforms include: 1) the acceleration of OpenAI enterprise business growth supports rapid annualized revenue operation rate (ARR) growth, and 3Q26's enterprise business ARR increased 107% year-on-year (according to CNBC); 2) advertising commercialization continues to advance. At the level of progress in commercialization of domestic AI platforms, the revenue sharing model for overseas cloud platforms has been successfully expanded.
Future outlook: RSI accelerates model iteration, reduces cost per task, increases AI penetration rate, and focuses on agent progress
CMB International expects that the continued deepening of RSI is expected to further increase the AI penetration rate. Anthropic revealed that the share of AI-led R&D work had risen to 26% in August (March: 1%). Increased model capability and reduced cost per task together improve user ROI. Improved computer operation capabilities enable AI to participate in more non-programming office processes. Examples of efficiency improvement in daily work are expected to have a positive demonstration effect, further improving the penetration rate and depth of use at the B and C ends. As agents progress and deepens, competition in the industry is expected to gradually shift from “who has the stronger model” to “who can have a long-term workflow for users”. There is a risk that the value of mid-tier SaaS and automation tools will be squeezed in the future, and governance and compliance are expected to receive more attention.
Recommended targets
1) Overseas cloud vendors: It is recommended to focus on Amazon (AMZN.US), Microsoft (MSFT.US), and Google (GOOGL.US); 2) Chinese cloud and AI applications: Alibaba (BABA.US) and Tencent (00700); 3) Big model targets: optimistic about Intellect (02513) and MINIMAX-W (00100).