Summary of Key Points
Following the release of GPT-5.6, the number of active users on OpenAI’s Codex and ChatGPT Work surged from 6 million to 9 million in just a few days, an increase of 1 million per day. Both companies have been competing for users by repeatedly resetting quotas and lifting usage restrictions. Their rival, Anthropic, has also adjusted its strategy for using its Claude model in response. However, OpenAI’s efforts to integrate the two products have caused confusion among users due to the blurred distinctions between their functions. This competition has essentially shifted from a focus on model capabilities to a battle for user habits and the optimization of product interfaces.
1. Millions of Users Added Daily, but Behind the Numbers Lies Uncertainty
The figures for active users on Codex and ChatGPT Work published by OpenAI are impressive, yet they lack clarity:
- Unclear Metrics: It is not specified whether “active” refers to daily or weekly usage, nor is it clear what actions constitute being active (for example, does just clicking on the interface count as active usage, or do users need to complete tasks?).
- Mixed User Base: The numbers include both former Codex users who have switched over and new ChatGPT users trying out the new features. It is impossible to determine exactly how many new users have been acquired.
- Real Driving Force for Growth: The enhanced capabilities of GPT-5.6, such as helping users organize their digital belongings or proving mathematical conjectures within an hour by scheduling 64 sub-agents, are the main reasons users are willing to give it a try. Additionally, OpenAI’s frequent quota resets allow users to use the services for free, which naturally attracts more people.
2. OpenAI vs Anthropic: A Contest of Generosity in Quotas
The two companies are engaged in a sort of “computing power subsidy war” to prevent users from switching to each other’s platforms:
- OpenAI’s Aggressive Moves: When the number of users reached 6 million, OpenAI removed the 5-hour usage limit on the Plus package, allowing users to use the entire weekly quota at once. As the user base grew to 7 and 8 million, it continued to reset quotas and even offered coupons for future quota resets.
- Anthropic’s Countermeasures: On the day GPT-5.6 was released, Anthropic extended the trial period for its Claude Fable 5 model and reset users’ quotas. It later extended the trial period until July 19th, offering a 50% higher weekly quota than usual for Claude Code.
- Social Media Battles: OpenAI’s mascot,奥特曼, shared an promotional video for Anthropic, joking that it seemed sarcastic and implying that Anthropic restricted user access. OpenAI also launched a “feedback for points” campaign, offering $100 in credits to the top 10,000 users who explained why they liked GPT-5.6—essentially, trying to win over users’ loyalty.
3. Two Interfaces That Are Hard to Distinguish? OpenAI’s Integration Confuses Users Even More
OpenAI aimed to merge Codex and ChatGPT Work into a single interface, but this has only added to user confusion:
- Blurred Functionality: While officially, Codex is described as a programming tool and ChatGPT Work as a tool for handling research and documents, in practice, both can read files, invoke tools, and execute long tasks. Users are unsure which one to use.
- Disruption of User Habits: Codex has evolved from a pure programming tool into an agent that can handle complex tasks, and many users have grown accustomed to using it for work. The sudden addition of ChatGPT Work, with its reduced chat functionality, has led to complaints; some users consider the change to “ChatGPT Desktop” a terrible decision.
- OpenAI’s Attempts to Improve the Experience: OpenAI later updated the search function in ChatGPT, allowing users to find past conversations and documents from the sidebar, aiming to make it a central hub for all work-related tasks—so users don’t need to worry about which interface to use and can simply continue with their work.
4. The Shift in AI Competition
The focus of competition among large AI companies has shifted from “who has the strongest model” to “who can create the most user-friendly experience":
- High Barriers to Switching: Once AI becomes integrated into users’ work processes (such as coding or research), it is difficult for them to switch platforms once they form a habit. Therefore, both companies are eager to encourage more usage.
- Architecture Competition Takes Precedence over Capability: OpenAI aims to create a unified interface for AGI, but it has encountered user experience issues with its integration (multiple interfaces and unclear functionality). Anthropic, on the other hand, relies on its “safe and reliable” brand image to retain users.
- The Ultimate Test of Practicality: Regardless of the amount of quota offered or the design of the interfaces, the ultimate factor determining user retention is whether the AI can truly help with tasks. If an AI agent is merely ornamental, even generous quotas won’t be enough to attract and retain users.
This competition marks a transition in the development of AI products from being technology-driven to being user-driven. The company that successfully packages complex AI capabilities into intuitive and useful products will gain an advantage in the era of agents. OpenAI’s and Anthropic’s “quota wars” are just the beginning of this battle.