OpenAI Dev Day Unveils 25 Updates: Agent dots Debuts, ChatGPT Memberships Shift to 'AI Employee' Pricing

In the early hours of September 30 Beijing time, OpenAI held what it called its 'largest-ever' developer conference, releasing 25 updates at once, covering new models, APIs, agents, the plugin ecosystem, and an enterprise software marketplace ^[1][2][3].

Membership Changes: Cheaper Models, but Plan Quotas Are Shrinking

The quietest moment on stage came when Sam Altman discussed ChatGPT Pro membership adjustments—according to descriptions, not a single person in the audience applauded ^[1][2].

The adjustment itself is not complicated: the $200/month Pro 200 price stays the same, but new subscribers' quota is halved; at the same time, a new $500/month Pro 500 tier is added, with 25 times the quota of Plus, and it exclusively offers Astra Ultrafast high-speed mode, which only it can use among personal plans ^[1][2].

In contrast to the tightened plan quotas, API prices for new models such as GPT-6.1 Sol were significantly lowered, but usable quota within membership plans was simultaneously reduced, triggering developer dissatisfaction ^[1].

dots: An Agent Priced Like an 'AI Employee'

The highlight of this conference was the agent dots, powered by GPT-6 Astra. According to the official description, each dot has its own cloud computer and browser, can connect to thousands of apps (officially stated as more than 4,000), runs around the clock, and can continuously complete tasks autonomously ^[1][2].

Usage is close to 'assigning work': give it a goal and define which operations it can decide on its own, and it will proactively start working; it can also stay in ChatGPT, Slack, and Teams. OpenAI product lead Tibo said on stage that you can even call it directly while it is working ^[2]. The official vision is that multiple dots can form a team to work collaboratively in the future ^[2].

The threshold is not low. Among individual users, only Pro plans can use it, including the $100/month Pro 100, while on the enterprise side it is Business Premium; officials say the first dot is included with the plan and chatting with it does not count toward usage, but sending it to work in ChatGPT workspaces and Codex still consumes quota ^[1][2].

Along with dots comes a shift in pricing logic: OpenAI no longer simply charges by number of tokens, but instead references an 'employing AI employees' model, charging separately by delivered workload and operating speed; high-speed mode consumes quota faster ^[1][2].

Platformization: Plugins, Accounts, and Cross-App Quotas

OpenAI is simultaneously advancing platformization, adding a plugin ecosystem, ChatGPT account login, and cross-app universal quota capabilities, benchmarking WeChat mini programs and account systems, attempting to turn ChatGPT into a unified entry point for collaboration between people and agents ^[1][2].

For enterprises, OpenAI launched team data spaces, dynamic documents, a new version of the Codex cloud development environment, and a software marketplace, where enterprise quotas can be used to directly purchase third-party commercial software ^[1][2].

Implementation still raises questions. At a Silicon Valley livestream viewing event, one attendee working in mathematics mentioned that his code needs to call a mathematical optimization solver with a commercial license: local Codex can call, run, and test it, but after migrating to the cloud, these tools and licenses may not follow ^[2].

Also Announced: DeepSeek Harness v0.2 Preview

On September 29, DeepSeek Harness v0.2 Preview was released, offering out-of-the-box macOS and Windows desktop installers, and it is now available on the DeepSeek official website. The new version comes with preset common features for daily office work and development, adds a plugin installation and management page, improves the settings and management of automated tasks, and optimizes file display, content preview, and code change display experiences ^[1].

Officials say users can provide documents, spreadsheets, or PDF files and have it organize materials, analyze data, generate charts, or create presentations; they can also have it modify projects and run code. The team says it will continue to improve the experience from onboarding to task completion, including improving sandbox features and security, and refining agent team features ^[1].