The $20 AI coding tier is quietly collapsing

Three moves in four days. Taken together they tell one story:

April 20. GitHub paused new signups on Copilot Pro and Pro+ individual plans, tightened monthly usage limits, and removed Anthropic’s Claude Opus 4.7 from Copilot Pro. Opus access is now Pro+ only. Users hitting new limits can cancel and get a prorated refund through May 20. GitHub’s explanation: agentic workflows “regularly consume far more resources than the original plan structure was built to support,” and weekly operating costs roughly doubled starting January 2026.

April 21. Anthropic quietly removed Claude Code from the $20/month Pro tier on new-signup pricing pages. After several hours of community backlash, Anthropic reversed the change and characterized it as an A/B test touching 2% of new signups. Existing Pro users kept Claude Code. The trial balloon was unmistakable: how far can we push Claude Code upmarket before users revolt?

April 16–21. OpenAI shipped Codex for (almost) everything — macOS Computer Use, 90+ plugins (Atlassian, CircleCI, GitLab Issues, Microsoft Suite), SSH to remote devboxes, memory/personalization. Weekly active developers grew from 3M to 4M+ in under three weeks. On April 21, OpenAI launched Codex Labs and GSI partnerships for enterprise rollout, plus Chronicle — an on-device screen-capture memory feature for macOS Pro users. Codex is now firmly aimed at the Pro ($100) and Pro $200 tiers, not at $20 Plus users.

What these three moves have in common

Agentic coding — the workflow where an AI reads your codebase, plans multi-file changes, runs tests, iterates until done — consumes 10-50x more tokens than autocomplete. The economics that worked when “AI coding” meant “suggest the next line” don’t work when it means “spend 40 minutes refactoring an auth system.”

The $20 tier was designed for the old workflow. Every vendor is now quietly updating its tier structure for the new one. Copilot was most direct about it — openly citing “doubled weekly operating costs.” Anthropic tried to do it silently and got caught. OpenAI is doing it by making Codex’s best features (Computer Use, plugins, Chronicle) available to everyone but practically unusable at Plus-tier credit budgets.

What this changes for you

If you pay $20/month for Copilot Pro today: Opus 4.7 access is gone. If Opus was your main reason for the subscription, you’re effectively on a different product now. Consider moving to Copilot Pro+ ($39) or switching to Claude Code via Claude Pro ($20).

If you pay $20/month for Claude Pro for Claude Code: You kept the product this week, but I wouldn’t bet on it lasting. Heavy Claude Code users have been migrating to Max 5x ($100) for six months because Pro caps cut them off daily. The pricing page A/B test confirms the direction.

If you pay $20/month for ChatGPT Plus and use Codex: The best new Codex features — Computer Use, Chronicle, many plugin integrations — are gated to Pro tiers or have such low Plus usage caps that they’re demos, not tools. If Codex is central to how you work, $100 Pro is now the realistic floor.

Near-term tradeoffs:

• Claude Pro ($20) + Claude Code: Still the cheapest way to get Opus-caliber agentic coding in April 2026. Possibly not for long.
• Cursor Pro ($20): Agent mode uses Cursor’s own Composer 2 by default; Opus access is via BYOK or higher tiers. Less exposed to the upstream pricing squeeze because Cursor owns its model.
• Copilot Pro+ ($39): Currently the cheapest way to get Opus-on-Copilot. If you want GitHub-native agent work, this is now the entry point.
• Codex via ChatGPT Pro ($100): The new reference tier for serious Codex agent work. Paying $100/month felt insane six months ago; it’s becoming normal.

The contrarian take

None of this is bad. The old $20 tier was quietly subsidized by the fact that most users didn’t push the tools. As agentic workflows go mainstream, the tools actually cost what they cost, and pricing catches up. The transition is uncomfortable for solo devs and hobbyists. But the alternative — vendors running at a loss until someone goes bankrupt — is worse.

What I’d watch next:

1. Does Anthropic try the Pro/Claude-Code removal again in two months? I’d bet yes.
2. Do Plus-tier Codex caps tighten? Watch for “credit” language creeping into the Plus product page.
3. Does Cursor stay at $20 longer than the others? It has the strongest reason to (owns its model, doesn’t pay Anthropic per-token).

What I’d tell a developer asking today

If you’re spending significant time doing real agentic coding work and you’ve been on the $20 tier of anything, the next subscription cycle is the right time to upgrade deliberately. Paying $100 once beats paying $20 three times and getting rate-limited mid-project.

If you’re a hobbyist who uses AI for suggestions, nothing changes this week. But watch the next 60 days — I expect the Plus/Pro line to keep tightening across every vendor.

The $20 era of agentic coding was a beautiful anomaly. It’s ending. Plan for it.

Questions or corrections? Email Pick Right. Want the full list? See all news.