-+ 0.00%
-+ 0.00%
-+ 0.00%

CITIC Construction Investment: Continued upgrading of flagship models and agent products is expected to further enhance commercial value

Zhitongcaijing·11/17/2026 23:57:06
Listen to the news

The Zhitong Finance App learned that CITIC Construction Investment released a research report saying that the flagship model and agent products continue to be upgraded, and they continue to be optimistic about leading platforms with both models, cloud resources, and application portals. Qwen 4 Max and Claude 5.2 enhance programming, professional tools, and long-task capabilities. Google complements enterprise reasoning systems, OpenAI launches merchant services and authorized payments, and AI applications are expanding to specific business operations. From November 9 to 15, OpenRouter calls increased by 0.86% month-on-month; HY4 and MIMO V2.6 Flash increased by 41.61% and 87.91% respectively; Claude Code's 7-day average download volume increased 25.37%, and WorkBuddy store revenue increased 62.12%. Model upgrades and improved deployment products are expected to drive enterprises to increase calls and payments, and platforms with advanced model capabilities and mature cloud delivery will benefit more fully.

CITIC Construction Investment's main views are as follows:

Platform calls are growing steadily, and usage of Hy4 and Xiaomi models is increasing

From November 9 to 15, OpenRouter's weekly usage was 211.28 trillion tokens, up 0.86% month-on-month. Hy4 usage was 18.12 trillion yuan, an increase of 41.61%, and MiMO V2.6 Flash usage was 4.23 trillion yuan, an increase of 87.91%. The total number of platforms is stable, and the adoption of some domestic models is improving rapidly.

pictures

Increased adoption of programming tools, Claude Code leads the way in download growth

As of November 14, Claude Code had an average of 2.52 million downloads per day, an increase of 25.37% over a week ago; Codex was 3.79 million times per day, an increase of 1.88%. Kilo Code and Codex calls on the OpenRouter side have also increased, and programming is still the direction where agent requirements are more clear.

pictures

WorkBuddy's usage and payments both increased, and revenue increased by a thousand questions

WorkBuddy had 918,900 weekly downloads, an increase of 17.91%, store revenue of 612,000 US dollars, an increase of 62.12%; Qianwen Store revenue increased 39.92%. Gemini downloads increased 24.01% in the US. Application performance is fragmented, and the use and payment of office tools has increased significantly.

pictures

The flagship model accelerates iteration, and the ability for complex tasks continues to increase

Ali released the official version of Qwen 4 Max on November 16, supporting native vision and 1 million token contexts; Anthropic released Fable 5.2 and Mythos 5.2. Leading manufacturers continue to improve programming, professional tools and long-term task capabilities, and model competition penetrates deep into the actual work processes of enterprises.

The share of open source models has rebounded, and commercialization opportunities cover a wide range of models. On November 15, the Vercel AI Gateway open source model token share was 72.57%, up 1.50 percentage points from a week ago; on the spending side, GLM-5.3's share rose to 12.92%, up 4.88 percentage points, and Kimi K3.1 Pro's share increased 2.10 percentage points. Claude Opus 5.5 is still in an important position on the spending side with a 23.09% share, and the newly released Fable 5.2 already has a 5.77% share. The domestic model is progressing on both the billing and expenditure sides, and the new closed source model is also rapidly gaining payment requirements. Companies are more specific in choosing models, and there are different markets for cost efficiency and complex task capabilities. Vendors that continue to iterate and can enter actual workflows have an advantage.

Long context and security requirements drive the continuous upgrading of cloud inference systems. Google added TPU 8tX and TPU 8i Max systems: the former is based on the TPU 8t and combined with a fifth-generation AMD EPYC processor to provide hardware protection for inference models and data, and is aimed at sensitive businesses such as finance and government; the latter integrates two TPU 8i in the same package to increase computing, on-chip storage and memory bandwidth, and support ultra-long contexts and multi-agent tasks. Increased model capabilities are bringing more complex deployment requirements. Cloud platforms launch specialized products around security, latency, and long-term task processing, which can help meet the needs of enterprises from trial to production deployment. A platform with model services, self-developed computing power, and enterprise customer base, which is expected to expand the cloud business in this process.

GPU rental price performance is divided, and high-end reasoning resources are still supported by demand. As of November 14, the H100 rental price was $3.19 per GPU hour, up 2.90% from a week earlier; the H200 was $6.60 per GPU hour, up 0.46%, and once rose to $7.75 per GPU hour during the week. B200 fell back to $9.66 per GPU hour, down 2.23%, A100 fell 1.92%, and RTX 5090 remained flat. Prices of different models have not risen at the same time, and the rise in H100 and H200 indicates that demand for some high-end resources remains resilient. Combined with Google's new system for long contexts and confidential reasoning, computing power products are being further segmented around specific tasks, making it easier for cloud service providers that can provide appropriate resources and improve deployment efficiency to handle incremental demand.

Merchant services and authorized payments complement the transaction process, and agent applications expand to business scenarios. OpenAI launched a beta version of the ChatGPT merchant service. Merchants can maintain data, price, and availability through the Business Hub, and eligible US merchants can also configure Business Agents to answer product and service questions; ChatGPT wallets are gradually being opened to eligible US Pro adult users to support credit card payments after user authorization. The access to merchant information, product Q&A, and payment functions enables ChatGPT to undertake more transaction-related tasks, and also provides a direction for platforms with user portals to expand merchant services. As tool calls and long-term task capabilities increase, the scope of services that agents can serve is being expanded from content generation to specific business operations, and the commercial value of application portals is expected to further increase.

Continue to be optimistic about model capabilities, cloud services, and application portals to form a collaborative leading platform. In this week's high-frequency data, the total number of platform calls increased slightly, and the increase in programming tool downloads and payments for some office applications was even more prominent; on the product side, Qian Q and Claude continued to improve complex task capabilities, Google added enterprise reasoning systems, and OpenAI extended merchant and payment functions. Increased demand and product upgrades are mutually driving each other in specific scenarios. In terms of investment, the focus is on platforms that can continuously launch high-capacity models, accept enterprise calls through cloud services, and expand payment scenarios based on application portals. Among them, programming and office agents have made clear usage and payment progress, which is an important direction for observing AI commercialization at present.

Risk Alerts

Technology iteration and competition increase risk; risk of commercial monetization falling short of expectations; risk of computing power delivery, cost and return on capital; risk of content security, payment security, and regulatory risks; and risk of high-frequency data tracking and estimation bias.