AI系列34：云服务商涨价；快乐马；M2.7开源；GLM-5.1发布

Highlights (1)CSPAliCloud and Baidu will raise prices on 18 Apr whileTencent Cloud on 9 May amid surging demand amid supply constraints;(2) BABA’sHappy Horse is a surprise and ranked at top position invideo generation model; (3)MiniMax M2.7is open source starting on 12Apr; (4) Zhipu launchedGLM-5.1model and raised prices last week; (5)OpenRouter: weekly tokens consumption down vs prior week due to end ofpromotions for some models on its platform. In this report, we provide Investors with over 50 data points on AI models, OpenClaw, user trendsby sub-sector, and other industry data for analysis. CSP will start raising prices later this week on surging demand amid supply constraints.AliCloud (18 Apr) :(1) AI computing (e.g. Zhenwu 810E) increases by about 5% to 34%; (2)CPFS storage increases by about 30%;Baidu (18 Apr): (1) AI computing power related productsby 5% to 30%; (2) parallel file storage and others increases by 30%;Tencent Cloud(9 May)(1)AI computing, container services and Elastic MapReduce up by 5% BABA’s video generation model Happy Horse is unexpected surprise.It ranked first in text tovideo leaderboard in Artificial Analysis. It is under testing and will open its API in near future.Separately, AliCloud will stop Coding Plan Lite version on 13 Apr and users can buy Pro versionafterwards. MiniMax’s M2.7 model is open source starting from 12 Apr.It partners with domestic andoverseas chips companies on model integration and inference adaption. We view open sourcemodels can strengthen developers community and branding. It does not affect enterprisesusing API (whether close or open source) as self deployment cost is relatively high. MiniMaxachieves low costs through inference engineering. ARR is determined by intelligence ratherthan close/open source (M2.1/2.5 are open source). Zhipu launched GLM-5.1 model and captures top positions a number of assessment.Theopen source model was released on 7 Apr with further enhancements in coding and agenticAI capabilities (e.g. it ranked no 3 in SWE-Bench Pro, Terminal-Bench 2.0 and NL2Pepo andbehind GPT5.4 and Opus 4.6). Further price hikes by about 10% seen across models last week. OpenRouter: Token consumption was down by 23% vs prior week.Based on the analysis ofthe week of 6 Apr (vs the week of 30 Mar) (Exhibit 1), number of weekly tokens consumeddecreased by 23% to 21tn. We believe this is due to (1) OpenRouter has promotional campaignsfor Mimo-V2-Pro model which ended on 2 Apr; (2) OpenRouter has promotions campaigns forQwen3.6 Plus (free) between 2 to 16 Apr. There are limitations set on free API calls after 6 Apr.In terms of ranking, BABA's Qwen3.6 Plus (free) ranked no. 1, followed by DeepSeek V3.2 andClaude Opus 4.6 BAT, KC, and AI Labs are Beneficiaries.We expect surge in token consumption on risingdemand in AI models and agents to benefit CSPs likeBaidu, Alibaba, Tencent (BAT), KingsoftCloud (KC), and AI Labs.According to analysis from Artificial Analysis, (1) there is a narrowingintelligence gap between US and China models; (2) API cost is just a fraction vs US benefitingfrom MoE, linear attention, and MFU (Exhibit 22-28); (3) Coding, Workspace scenario and mid-to long-form content have significant opportunities ahead (Exhibit 53-57). Continued overleaf... China vs US comparison Input/Output Price for Different Models / Pricing forCoding Plan and Agents Offerings Coding Plans and Agents offerings AI Models According to OpenRouter, rising adoption ofOpenClaw and release of high-performanceand cost-efficient China models drive tokenconsumption According to OpenRouter, token consumption went up significantly in Feb-26 (a)surge in adoptionof open-source OpenClaw, which allows people using AI from "to chat" to "to work". It connects withlarge models which can be deployed locally or on servers which allows you to automate workflowthrough natural language chat; (b)release a number of high performance and cost-efficient modelwhich includes Kimi K2.5 (Jan-26), M2.5 (Feb-26) and GLM-5 (Feb-26). According to OpenRouter (between 6 Apr and 12 Apr), China models token consumption decreasedby 55% compared to prior week reaching 5.8tn. We view this is due to (1) OpenRouter haspromotional campaigns for Mimo-V2-Pro model which ended on 2 Apr; (2) OpenRouter haspromotions campaigns for Qwen3.6 Plus (free) between 2 to 16 Apr. There are limitations set onfree API calls after 6 Apr. This compares to US models at 4.1tn token consumption. China models surpass US models again interms of token consumption the week of 6April Qwen3.6 Plus (free) ranked first in terms oftoken consumption the week of 6 April Based on Artificial Analysis, China models DeepSeek, Qwen, Kimi, MiniMax and Zhipu are closingthe gap with US models in terms of intelligence. China models are highly cost-efficient on Mixtureof Experts (MoE) architecture, linear attention, and MLU (Model Flog Utilization). China models are narrowing the gap with USmodels in terms of intelligence. On Input