What this is
We noticed a high-engagement post on Juejin where "Lao Jin" wrote that when the Claude Opus 5.5 and GPT-6 series (Sol, Luna) launched in the early morning hours, his first reaction wasn't to top up on the official site—but to check whether his Agent.Space subscription balance would survive a new round of usage.
Agent.Space is a third-party AI aggregator that has recently emerged in China. Billing comes in two flavors: Share is a fixed-quota GPT package at roughly half the official price; Flex is a usage-based universal credit covering Claude, Grok, DeepSeek, Kimi, GLM, MiniMax, and other models. Monthly pricing is Plus at $10 and Pro 5x at $50; annual billing gets an additional 10% off, with payment routed through domestic channels.
The differentiator is the Workspace (collaborative workspace): it puts files, conversations, previews, and task progress into one project, letting multiple AI assistants (Agents) take turns—Codex handles the page, Claude Code reviews code and copy, image generation directly calls GPT Image, and results stay in the workspace. It also supports a Developer API Key, so you can take your quota back to local command-line clients.
Industry view
The article is essentially a plug, but the phenomenon it reflects is real and worth our attention: the budgets of China's AI heavy users are systematically flowing to third parties. The reasons aren't complicated—frequent bans on official accounts, high barriers to official APIs in China, and the constant flight of relay services. These three forces together have created strong demand for "reliable relay + half price + one-stop Workspace."
Counterarguments are worth hearing too. First, the discount source is opaque—is it subsidies, time-zone arbitrage, or gray-market workarounds for official risk control? If the latter, compliance risk passes through to users. Second, Workspace is a good story, but the real moat in cross-agent collaboration is data sovereignty and permission governance—the article doesn't explain how each Agent's context is isolated. Third, for enterprise IT, "bypassing official channels" is itself a major compliance and auditability problem; half price doesn't equal low risk.
Others argue these platforms fill a gap in China's AI application layer—abroad have Cursor and Replit and other integrated products, but China has never had an equivalent-tier player. Agent.Space and its peers may not scale, but they're testing paths no one abroad has tried.
Impact on regular people
For enterprise IT: before choosing a third-party aggregator, ask three questions—can call logs be exported? Does data pass through third-party servers? Who bears compliance liability in the contract?
For working professionals: if you only occasionally use AI for emails or summaries, official channels plus an occasional proxy is enough; only when your workflow already stably depends on multi-model, multi-agent collaboration do the few hundred yuan saved monthly and the fewer tabs switched justify switching platforms.
For the consumer market: the real signal here is that AI models' "distribution layer" in China is decoupling from OpenAI and Anthropic and falling into the hands of these aggregators. Models increasingly look like utilities (water, electricity, gas)—the money is in the distribution and service layers.