Author on $200 Claude Plan Goes Back to 200K Context to Stay Within Limits

In spring, a $100 Claude subscription covered every task in the Habr author's side project. By early summer he was regularly using up the entire five-hour token limit. By August, the quota covered only technical design work, and he had to wait until the following week to start implementation. Once a task was running, it ate the weekly limit in about three days. In September he moved to the $200 plan, but even there the limits didn't feel out of reach.

The Path from $100 to $200, and a Dead End

The author notes that friends and colleagues have long been on the top Claude and GPT plans at the same time, and some plan to add a third subscription. He called that path a dead end: spending grows with appetite, and no safety margin ever appears. So he started optimizing his token consumption.

The main conclusion in the available excerpt is unexpected. The cause is not the complexity of the tasks or the amount of code read. The author calls it "mundane everyday convenience." The Habr text breaks off at this point, and further details are hidden behind a "Read more" link. But the headline ("Why I Went Back to a 200K Context Window") shows what decision he reached.

Why Window Size Hits Your Limits

What follows is general LLM mechanics, not a summary of the article. In a conversation with a model, and even more so in agentic coding, every new request carries the accumulated history: the conversation, files already read, command output. The larger the window and the less often it is cleared, the more tokens each step consumes. Prompt caching softens the blow but does not cancel it. The long "tail" of context still counts toward the quota.

The convenience here is obvious. A wide window means not having to think about boundaries: no new sessions, no rebuilding context, no compressing history. The habit of accumulating everything in one session works until limits become the main constraint. At large volumes, this carelessness turns out to be the most expensive line item.

Key Figures from the Story

• Spring: the $100 subscription covered all of the project's tasks.
• Early summer: the five-hour token limit is used up in full.
• August: the limits cover technical design, and implementation is pushed to the next week.
• Launching a task: the weekly quota burns out in about 3 days.
• September: the $200 subscription, with limits still noticeable.
• Decision: return to a 200K context window.

Practical Takeaways

The story is useful for anyone hitting subscription limits on Claude, GPT or other agentic tools. Before paying for the next tier, it is worth checking your own habits for working with context:

• Split work into separate sessions instead of one endless conversation.
• Don't keep files and logs in context that the next step no longer needs.
• Separate the design stage from implementation: they need different sets of data.
• Compare token consumption at a 200K window and at an extended one, on the same tasks.

The author concluded that the problem is not the plan but how it is used. For the industry, this is a signal: larger context windows and higher subscription prices remove technical limits, but token economics remain. A user paying $200 still counts every token, and context management tools are becoming as much a part of the workflow as choosing a model.