OpenAI doubled Codex limits, Anthropic killed the 5-hour window a day later. The fight is over agent-hours
Within 25 hours, the two biggest coding-agent vendors both changed how long your agent is allowed to run, and neither touched the price.
OpenAI moved first. On October 29 Thibault Sottiaux, who leads Codex, posted: "We are doubling all usage limits for all Pro 100/200/500 tiers and Business Premium, effective today." The same post cancelled a cut that had been announced a month earlier: "We heard your kind feedback and won't be reducing usage limits for Pro 200." Plus is not included.
Anthropic answered on October 30. Max 5x is now just Max at $100 a month, and Max 20x becomes Ultra at $200. The real change is in the second paragraph: "We're also removing the 5-hour limit for verified Ultra accounts, starting today with most existing subscribers." Newer accounts keep a temporary limit until they are verified, which Anthropic says is about fraud and abuse. Weekly limits stay the same.
The 5-hour window has been the defining constraint of Claude Code for over a year. People planned their day around it, scheduled overnight runs to dodge it, and built tools to track it. For the top tier it is now gone, and only the weekly cap is left. That favors exactly the workloads the labs keep showcasing: long background runs and several agents in parallel.
There is a footnote on where the tokens actually go. A gist that reached 584 points on Hacker News says Claude Code re-sends each background agent's whole transcript to the model about every 30 seconds, just to write a 3 to 5 word status line in the agents panel. By its author's telemetry, that timer accounted for 16.4% of Claude Code spend, more than the main conversation at 15.7%. These are community measurements, not Anthropic's, but they show why a subscription quota feels smaller than it should. When limits loosen, harness overhead gets hidden. When they tighten, it is the first thing to cut.
Our read: subscriptions have become the battlefield, and the unit is agent-hours, not model quality. Read your limits the way you read cloud pricing, and look at what the harness spends on your behalf.
OpenAI announcement: https://x.com/thsottiaux/status/2115850175868727621 Anthropic announcement: https://x.com/claudeai/status/2116220869479428500 Plans: https://claude.com/pricing Status-line gist: https://gist.github.com/flamerix-ai/c3905705cc4205e5de3011fa2e74efce
← Back to all articles
OpenAI moved first. On October 29 Thibault Sottiaux, who leads Codex, posted: "We are doubling all usage limits for all Pro 100/200/500 tiers and Business Premium, effective today." The same post cancelled a cut that had been announced a month earlier: "We heard your kind feedback and won't be reducing usage limits for Pro 200." Plus is not included.
Anthropic answered on October 30. Max 5x is now just Max at $100 a month, and Max 20x becomes Ultra at $200. The real change is in the second paragraph: "We're also removing the 5-hour limit for verified Ultra accounts, starting today with most existing subscribers." Newer accounts keep a temporary limit until they are verified, which Anthropic says is about fraud and abuse. Weekly limits stay the same.
The 5-hour window has been the defining constraint of Claude Code for over a year. People planned their day around it, scheduled overnight runs to dodge it, and built tools to track it. For the top tier it is now gone, and only the weekly cap is left. That favors exactly the workloads the labs keep showcasing: long background runs and several agents in parallel.
There is a footnote on where the tokens actually go. A gist that reached 584 points on Hacker News says Claude Code re-sends each background agent's whole transcript to the model about every 30 seconds, just to write a 3 to 5 word status line in the agents panel. By its author's telemetry, that timer accounted for 16.4% of Claude Code spend, more than the main conversation at 15.7%. These are community measurements, not Anthropic's, but they show why a subscription quota feels smaller than it should. When limits loosen, harness overhead gets hidden. When they tighten, it is the first thing to cut.
Our read: subscriptions have become the battlefield, and the unit is agent-hours, not model quality. Read your limits the way you read cloud pricing, and look at what the harness spends on your behalf.
OpenAI announcement: https://x.com/thsottiaux/status/2115850175868727621 Anthropic announcement: https://x.com/claudeai/status/2116220869479428500 Plans: https://claude.com/pricing Status-line gist: https://gist.github.com/flamerix-ai/c3905705cc4205e5de3011fa2e74efce
Comments