Bilingual primary-source research · Shanghai
A source-checked release packet for AI evaluators, founders and technical publications—before you benchmark, publish or decide.
Foreign AI coding tools should test cross-border teams, private deployment and bring-your-own-key workflows before betting on China’s individual subscription market.
The job
English-language coverage tells you what travelled. We check what the original Chinese sources said, what local users are doing, and where those two stories disagree.
Independent research desk
The desk behind the memo
An independent bilingual research desk based in Shanghai. Each sprint traces product, pricing, policy and developer claims through Chinese and English primary sources, labels uncertainty, and keeps the answer small enough to use.
Public sample · checked 13 July 2026
Price floor
Distribution
Entry route
Recommended test
Recruit three China-based teams that collaborate with overseas clients or repositories. Offer a four-week pilot around bilingual issue-to-code execution, private/BYOK access and auditable data handling. Delay broad consumer spend until at least two teams agree to keep paying.
Market-entry research only. This sample is not legal or investment advice.Evidence appendix
Live product pages, documentation and official rule text. Accessed 13 July 2026.
Core functions are free; paid personal tiers sell queue priority and more concurrent cloud tasks.
Decision useAccess alone has a zero-price anchor. A paid entrant needs measurable capacity or workflow advantage.
Frequent 2026 releases added subagents, browser use, hooks, skills, mobile collaboration and Douyin login.
Decision useA foreign tool competes with a fast-moving product that already has domestic identity and distribution.
SaaS/VPC deployment, enterprise-owned models, usage analytics, cost limits and code-data assurances are offered.
Decision useDeployment, governance and observability—not generic chat—form the enterprise buying layer.
Alibaba states that individual use is free while the agent can inspect repositories, edit files and run commands.
Decision useA feature-rich domestic alternative already gives individuals a credible free default.
Team and enterprise plans package credits, knowledge controls, VPC and compliance features per user.
Decision usePayment demand is clearest around teams and organizational control rather than bare individual access.
Personal use is free; paid tiers add authorization, productivity reporting and dedicated deployment.
Decision useA second major vendor reinforces the free-personal, paid-organization pattern.
The agent can route across Alibaba, DeepSeek, MiniMax, Z.AI, OpenRouter, proxies and local providers.
Decision useAgent shells and model routing are commoditizing; reliability and proprietary workflow context must carry value.
The open model supports local deployment and common IDE integrations alongside repository-level tasks.
Decision useLocal operation raises privacy expectations and weakens model access as a standalone moat.
Official live prices keep domestic inference inexpensive, although they can change quickly.
Decision useBYOK can make pilots economical while bundled-model subscriptions face margin pressure.
Mainland model access is sold at low per-token prices and includes introductory trial capacity.
Decision useDomestic BYOK and local routing are more credible than forcing a foreign inference bundle.
The measures expressly address generative services offered to the mainland public and overseas noncompliant services.
Decision useA public launch is a regulatory product launch; a closed organizational pilot is a narrower first test.
The rules cover AI-generated text and define visible and metadata labeling obligations.
Decision useDisclosure and export behavior need product review before broad public availability.
What arrives
One question. One evidence packet. No deck theatre.
Official pages, filings, pricing, local reporting and developer evidence.
The answer first, contradictions surfaced, and what the evidence changes.
One narrow decision question, delivered without a calendar obstacle course.
A clean correction pass if a source, translation or factual claim needs repair.
How it works
A company, market, user behaviour or claim you need to verify.
Official pages, local pricing, product changes, user evidence and contradictions.
A concise English memo, evidence cards, links and a practical next move.
Founding-client pilot
Public-source research, completed asynchronously from Shanghai.
China Signal Sprint
$299fixed · scope confirmed before paymentOpen the desk
A fit check comes first. If the question is too broad, too private, or needs licensed legal advice, it does not become a paid sprint.