OpenAI vs Anthropic: The Agent War of Attrition

新智元 · wechat · 2026-07-18

OpenAI and Anthropic are fighting an "Agent War of Attrition"

This article analyzes the recent, almost simultaneous moves by OpenAI and Anthropic to relax/reset usage limits, suggesting both companies are vying for real-world, long-task data through higher usage caps and looser access strategies. This isn't just subsidizing users; it's a battle for the gateway of "who can get more actual work done" in the agent era.

Core Perspectives

On OpenAI's side, Sam Altman posted that the past 12 months were "less than ideal," but the next 12 will be "the best ever." This is interpreted as the company betting on more thorough paradigm shifts rather than just releasing a single new model.
OpenAI's Codex / ChatGPT Work usage has grown rapidly, with active user counts surging short-term, prompting the company to temporarily lift or reset usage limits to stabilize the system.
Anthropic concurrently extended paid access for Claude and bumped Claude Code's weekly limits by 50%. Both companies are "giving away limits to grab usage."
The article cites OpenAI CFO Sarah Friar's new perspective: AI metrics shouldn't just look at token costs, but at "how much useful intelligence per dollar"—the total cost of completing a task.

Fighting for Workflows, Not Just Seats

The article emphasizes that the core value of agent products isn't chatting, but continuously working for users for hours: breaking down tasks, calling tools, reading files, modifying code, and running processes until the job is truly done.

Under this framework, the core competition between OpenAI and Anthropic is:

Who can first habituate users to "letting AI do the work";
Who can acquire more real-world task data;
Who can transform AI from a "question-answering tool" into the "work method itself."

Conclusion

The author believes the outcome of this competition may not depend on model parameters, but rather on who first embeds into user workflows and becomes a true AI coworker.

Original post →

OpenAI vs Anthropic: The Agent War of Attrition

More from Companies & People