The most-watched countdown in AI expires tonight at 11:59 PM Pacific — and this time there’s an answer instead of an extension. Saturday’s betting was split three ways: Opus 5 ships, the deadline slips a fourth time, or billing simply starts. Anthropic picked a fourth door. Per the company’s announcement, reported by The Decoder, TechTimes and Dawn: starting July 20, Claude Fable 5 becomes a permanent inclusion in Max and Team Premium plans, capped at 50 percent of weekly usage limits. Pro and Team Standard users lose bundled access entirely and shift to prepaid credits — $10 per million input tokens, $50 per million output — cushioned by a one-time $100 credit.
The fine print is where it stings. Those “50 percent limits” are 50 percent of a smaller number: the same Monday, Anthropic ends the 50 percent boost to Claude Code weekly rate limits it has been running, so the base allocation shrinks just as Fable lands inside it. The Decoder’s read: heavy Pro users are being deliberately funneled toward API pricing, where the meter runs honestly. A $100 starter credit sounds generous until you do the division — at $50 per million output tokens, a serious daily workload can burn through it in days.
Step back and the six-week arc is remarkable: a government-ordered access pause that took Fable 5 dark for 19 days in June, a July 1 restoration, then three deadline slips in five weeks while Anthropic decided what its flagship should cost. And the resolution mirrors the whole industry’s move: OpenAI tiered GPT-5.6 into three models at three prices, and DeepSeek is bringing peak/off-peak token pricing to V4. Flat-rate access to the frontier is quietly being retired everywhere at once.
Our take: The $20-a-month-gets-everything era just ended at the frontier. Mythos-class compute doesn’t fit inside subscription economics, so Anthropic split the difference — keep the flagship as a retention perk for the $100-plus tiers, meter it for everyone else. That’s rational rationing, but it hands rivals a target: every Pro user staring at a credit meter Monday morning is a customer Grok 4.5 priced its launch around. The real tell will be whether “50 percent of shrinking limits” holds, or whether power-user churn forces a quiet re-expansion by August.
What to watch
- Opus 5. The “Opus 5” strings spotted in Cursor’s codebase suggest a successor is close — timed right, it absorbs the Pro downgrade as a handoff.
- Monday’s limit crunch. The Claude Code boost expires the same day Fable joins Max plans — watch developer forums for the collision.
- Credit burn rates. How long $100 lasts at $10/$50 per million tokens will define whether Pro users upgrade, meter, or leave.
- A counter-move. OpenAI bundling GPT-5.6 Sol deeper into paid ChatGPT tiers would be the obvious response.