中国AI模型监控（2026年8月）：升级浪潮、用量激增、云端上行

China AI Model Monitor (Aug): Upgradewave, Usage surge, Cloud upside Industry Overview Rapid model upgrade, Pricing hike, Cloud acceleratesIntroducingthe China AI Model Monitor: a visual, data-driven monthly recap of model 31 August 2026 EquityChinaInternet/e-Commerce intelligence, token usage & pricing trends. In Aug, we observed: 1) rapid model upgradecontinued for both cloud incumbents (Tencent Hy4 preview, Alibaba’s Qwen 3.8Max/Flash) and independent AI labs (Zhipu’s GLM5.3 Flash); 2) selected companies (e.g.DeepSeek) raised token price, which we attribute more to compute tightness rather thanpivoting from high-performance-low-cost strategy; 3) more companies launched low-pricing small size models (GLM5.3 Flash), offering more high-performance-low-costoptions; 3) a sharp increase of cloud platforms capex and cloud revenue acceleration inChina–2Q Capex: RMB53bn for Tencent, RMB68bn for BABA; 2Q cloud revenue: low-20% YoY for Tencent, 45% YoY for BABA. Keytrends in numbers Intelligence level:According to Artificial Analysis, Anthropic currently leads theIntelligence Index with Claude Opus 5, followed by Claude Fable 5 & OpenAI's GPT-5.6Sol. Kimi K3 is the highest ranked Chinese model/open-source model globally (95% ofOpus 5), followed by GLM-5.3 & Qwen 3.8 Max.Model usage:OpenRouter datasuggests Chinese models are leading for usage, with DeepSeek V4 Flash #1 followed byTencent’s Hy3. OpenCode data suggests strong adoption of lower cost open Chinesemodels, with DeepSeek V4 Flash dominating usage. Vercel data suggests DeepSeekgained usage share m/m and leads at 30% share, while Anthropic usage share wassecond (25%) and was first for spend (65%).Token pricing:As of August 26, LLMToken Expenditure Index was down 26% m/m but stabilizing at $1.07. DeepSeek raisedprice in August which we attribute to likely tight compute on hand.Capex:listedChinese cloud incumbents saw major capex increase in 2Q26–we now expect China top4 cloud platforms combined capex up 98%/44% YoY in 26E/27E. Forward-12mconsensus capex for Tencent/BABA was revised up by 25%/29% in the past 30 days. Alex Liu>>Research AnalystMerrill Lynch (Hong Kong)+852 3508 5090alex.liu3@bofa.com Miranda Zhuang, CFA>>Research AnalystMerrill Lynch (Hong Kong)miranda.zhuang@bofa.com Joyce Ju>>Research AnalystMerrill Lynch (Hong Kong)joyce.ju@bofa.com Joanna Du>>Research AnalystMerrill Lynch (Hong Kong)joanna.du@bofa.com Key sector/company events ahead and implicationsformajor AIcompaniesin China List of abbreviations can be found inExhibit 34 >> Employed by a non-US affiliate of BofAS and is not registered/qualified as a research analystunder the FINRA rules.Refer to "Other Important Disclosures" for information on certain BofA Securities entities that take responsibility for the information herein in particular jurisdictions.BofA Securities does and seeks to do business with issuers covered in its researchreports. As a result, investors should be aware that the firm may have a conflict ofinterest that could affect the objectivity of this report. Investors should consider thisreport as only a single factor in making their investment decision.Refer to important disclosures on page 20 to 22.13013426 Who’s who in the China AI model market IncumbentsMajor internet companies such as ByteDance (private), Baidu, Alibaba, and Tencent— many of which have established cloud businesses—have developed proprietary AIfoundation models in-house. Independent AI labsDeepSeek (private), MiniMax, Zhipu are independent AI labs comparable to OpenAI and Anthropic, distinct from established internet platforms. Exhibit1:AI model landscape isintensely competitivein ChinaAI model details, benchmark scores, and pricing comparison table Model Usage / Adoption In this section, we track global frontier models’usage and adoption on OpenRouter,OpenCode and Vercel AI and highlight Chinese frontier models’performance. For moreanalysis on US frontier model capability and pricing trends, please refer to our USteam’s report:BofA Frontier AI Data Tracker: Model rankings, usage data, and token/GPUcosts. Token Usage - OpenRouterModel usage trends from OpenRouter suggest strong adoption of lower-cost Chinese open models, with DeepSeek V4 Flash 0731 leading at 31.6tn tokens MTD, followed byTencent’s Hy3 at 26.2tn and Xiaomi’s MiMo-V2.5 at 19.1tn, as of August 17. Exhibit2:DeepSeek V4 Flash and Tencent Hy3 lead in token usageTop 15 Models by Token Usage (AugustMTD, trillion token) By providers, DeepSeek led the weekly token usage on OpenRouter at 20.3tn tokens,followed by Xiaomi’s 10.5tn tokens, OpenAI’s 8.8tn tokens, and Tencent’s 8.2tn tokens inthe week of August 17. Among the top 50 models, DeepSeek accounts for the largest token usage share at 22%,followed by Xiaomi at 11% and OpenAI’s 9% in the week of August 17. Compared to thesame week in July, the most notable share gains were DeepSeek (+4.7pt) and OpenAI(+4.6pt), while the largest declines were Xiaomi (down 7.8pt), A