Sponsored
Anthropic admitted it in early 2026: Claude Code users were hitting usage limits far faster than expected, and the company called the fix its top priority. The frustration is measurable. Effective 14 September 2026, the indexed weekly ceiling drops from 150 to 125 — a cut of roughly 17 percent.
Go to virgil-ai.cloud → Try it in the browser first
Two meters, one pool
Every paid Claude plan meters you twice: a session limit that resets every five hours, and a weekly limit across all models. Claude Code, Claude chat and Cowork all draw from the same pool. The result: even subscribers on the $200 Max plan report single prompts eating 10 to 20 percent of a daily allocation — and then a blocked screen until reset.
The alternative: pay by the task
A growing class of tools flips the model. Instead of a monthly subscription with invisible ceilings, you pay per inference — Cline, Aider and Sourcegraph ($5 minimum, no markup) among them. The newest entry, from Germany, is Virgil.
What Virgil is
Virgil is an AI agent that lives inside your terminal: it reads, writes, runs and fixes — real files, real commands, in your shell. No browser window, no context switch. You install it with one line:
No weekly cap, no five-hour lockout. You start with 60 free credits, then pay only for what it does — 1 credit = 1 cent, with the cost of every task shown plainly in the terminal. A Google account syncs your balance across terminal and browser.
Who it is for — and the limits
Virgil suits anyone who works in bursts and resents monthly ceilings. If you code all day every day, a flat subscription may still be cheaper — pay-per-use wins where usage varies. To be fair: Virgil is young, its ecosystem smaller than the incumbents'. But the core is here and it runs.
Go to virgil-ai.cloud → Try it in the browser first
This article is a paid advertisement. Editorial facts on Claude limits: as of September 2026.
Comments