On June 24, ByteDance dropped a pricing bomb on China's AI subscription market.

Doubao fully upgraded to the Doubao-Seed 2.1 LLM; daily Token calls exceeded 18 trillion; it supports hundreds of AI agents collaborating simultaneously, and integrated lifestyle services like ride-hailing. It simultaneously launched three paid subscription tiers: Standard 68 RMB/month, Enhanced 200 RMB/month, Premium 500 RMB/month.

For comparison: ChatGPT Plus is 135 RMB/month, Pro is 677 RMB/month. Doubao's price is roughly half or even one-third of ChatGPT, and free-user benefits are fully preserved -- no feature cuts, no model restrictions, no forced upgrade pressure.

Alibaba Cloud's "80% Off at Night"

The same day, Alibaba Cloud gave an even more interesting answer.

Desktop intelligence tool QoderWork launched China's first AI compute time-of-use pricing mechanism: every night from 10 PM to 8 AM, calls to the Qwen3.7-Max flagship model are as low as 20% of regular price. Heavy users' batch long-task costs can be cut by 60%-80%; the officially recommended usage pattern is: "edit prompts during the day, let AI run autonomously at night."

This is an imaginative pricing innovation. It means AI compute is moving from "unified pricing" to "peak-valley time-of-use," becoming a public utility like electricity. The power grid has peaks and valleys; compute does too -- expensive during the day when everyone's using it, cheap at night when machines sit idle.

Infrastructure for the Agent Era

As prices fall, infrastructure is also being rebuilt for the Agent era.

At its Compile developer conference, Cursor launched Origin -- a cloud-native Git platform designed specifically for AI agents. Traditional GitHub was designed for human programmers: Pull Requests, Code Review, merge conflict resolution, all based on the assumption that "humans will read the code." But when hundreds of Agents simultaneously read and write to codebases, traditional Git can't handle it. Origin solves exactly this: automatic merge conflict handling, automatic CI failure fixing, automatic rollback of problematic commits.

The AI video space is likewise undergoing a paradigm shift. HeyGen open-sourced the Hyperframes framework (writing videos in HTML); OpenMontage released the world's first open-source agentic video production system -- 12 pipelines, 52 integration tools, 500+ agent skills. Video production moves from manual editing to code-driven automation; one person plus a fleet of Agents can produce video content that used to require a team of a dozen.

Chinese AI's Commercialization Path

Doubao and Alibaba Cloud's pricing strategies reflect a unique commercialization path for Chinese AI.

American AI companies take a "premium harvesting" route: first serve enterprise customers and professional users willing to pay high prices, then gradually move downmarket. But the Chinese market logic is entirely different -- it has the world's largest internet user base, the fiercest price competition, and the most mature mobile payment ecosystem; "low-barrier user acquisition + tiered monetization" is the repeatedly proven way.

What's 68 RMB/month? A cup of bubble tea plus a takeout order. This pricing isn't competing with ChatGPT; it's competing with video subscriptions, music subscriptions, cloud storage memberships -- turning AI from a "professional tool" into a "national app."

More critically, it has signal value: Chinese consumer AI is moving past the "free burn for users" stage and starting to think seriously about making money. When prices hit one-third of ChatGPT and there's still profit margin, it means Chinese models' inference cost optimization has been taken to an extreme. That cost advantage will ultimately translate into global market competitiveness.

The Utility-ification of AI

68 RMB/month, 80% off at night, Agent-native Git -- these seemingly unrelated news items point to one trend: AI is moving from "luxury" to "public utility."

Just as electricity moved from factory-exclusive to affordable for every household, just as broadband moved from premium business to civilian infrastructure, AI compute is going the same way. When calling an LLM costs less than sending a text message, when Agents can work for you 24x7 at extremely low cost, true AI mass adoption will arrive.

Price slayers aren't a bad thing. They pull technology from ivory towers to ordinary homes, letting more people afford and use it. Behind Doubao's 68 RMB is a market of hundreds of millions of users, the foundation for large-scale Agent commercialization, and the unique competitiveness of China's AI ecosystem.

Write prompts during the day, let Agents work at night, and the electric bill is cheaper than coffee -- that day may be closer than we think.

See you tomorrow.