The Compute Brief · Aug 13, 2026

DeepSeek updates V4 Pro as Chinese compute buildout advances

DeepSeek dominated the day’s China AI compute news, while Tencent, China Mobile, Cambricon and Sichuan added signals on infrastructure spending, chip adaptation and regional compute planning.

• DeepSeek announced the official DeepSeek-V4-Pro update across its app, web platform and API, with Expert Mode access, native OpenAI Responses API support and targeted Codex adaptation.

• DeepSeek also released a v0.1 developer preview of Harness, an MIT-licensed open-source code-agent framework based on Cordis, alongside a plugin ecosystem, while a separate item said the V4 Pro official version was abruptly withdrawn.

• DeepSeek announced peak-valley API pricing effective 00:00 Beijing time on August 17, 2026, with off-peak prices set at half of peak-hour prices outside 9:00-12:00 and 14:00-18:00 Beijing time.

• Tencent’s AI compute push remained in focus, with reports citing a 52.8 billion yuan computing power bill, Q2 capex up 176% year on year, and a strategy of prioritizing model building while treating AI workload rental as a fallback.

• China Mobile reported first-half 2026 operating revenue of 538 billion yuan, down 1.1% year on year, while computing power services and intelligent services accounted for 22.6% of core business revenue, up 2.2 percentage points from a year earlier.

• Cambricon said first-half 2026 revenue rose 108.13% year on year to 5.996 billion yuan and net profit attributable to shareholders rose 122.61% to 2.311 billion yuan, with inference adaptation completed for GLM, DeepSeek, Qwen, Kimi and MiniMax.

• Sichuan issued trial policy measures to strengthen the Chengdu Plain computing power core area and build an integrated computing-and-power belt across Panxi and northwest Sichuan for non-real-time workloads such as model training, big data analytics, rendering and backup.

• Moore Threads was highlighted as a case study in commercializing Chinese GPUs after doubling first-half revenue, improving gross margin and sharply narrowing its net loss.