OpenAI and Anthropic tightened usage limits on their paid plans between March and April 2026, drawing complaints from developers using Claude Code and ChatGPT Codex that they hit their caps within short periods. A class-action lawsuit was also filed against Anthropic in June by subscribers to its top-tier plan.
March–April 2026 · The Agent Cost Crunch
The "Unlimited" Subscription Breaks Under AI Agents
OpenAI and Anthropic quietly tightened usage caps on paid plans, triggering developer backlash — and a class-action lawsuit. Even $200/month top-tier subscribers report hitting limits in under 90 minutes.
$200/mo
Top-tier plan still hit its cap in under 90 min
10–20×
Token consumption spike from a prompt caching bug
5-hr
Rolling window that felt "far shorter " in real work
Why the flat rate stopped working
A single human prompt is cheap. An autonomous agent loops — calling tools, running tests, generating code — burning orders of magnitude more tokens per task.
Many ×
Agent (loops & tools)
Autonomous agents like Claude Code & Codex consume tokens continuously, not once.
The Two Frameworks · Paid Plans
Anthropic · Claude
Pro $20 · Max 5x $100 · Max 20x $200
Limit: 5-hr rolling + peak-hour tightening (5–11 a.m. PT)
Extra: third-party agents via separate credits
OpenAI · Codex
Plus base · Pro 5x $100 · Pro 20x ~$200
Limit: 5-hr rolling + weekly cap
Extra: partial move to pay-per-token
Developer criticism
Cap hit in 90 min even on $200 Max 20x
Caching bug burned 10–20× normal usage
Opaque limit displays — hard to plan work
Class-action lawsuit filed in June
Practical workarounds
Work nights/weekends to avoid peak hours
API pay-as-you-go is more predictable
Upgrading to Pro improved things
Use "/clear" to reset conversation history
The open question of the agent era: how to reconcile flat-rate subscriptions with GPU shortages and workloads that consume tokens without limit.
Backdrop: a GPU capacity shortage, surging demand for Claude after an OpenAI user exodus, and agent workloads that make "unlimited monthly" unsustainable. Adjustments and complaints continued into July.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…