Claude Fable 5 Subscription Access Settles Into A Two Tier Split

Anthropic has ended six weeks of rolling deadlines around Claude Fable 5 subscription access, confirming on Saturday that its most capable model will be included in all Max and Team Premium plans from 20 July, capped at 50 per cent of those plans’ usage limits. Pro and Team Standard subscribers keep the model only through metered usage credits, cushioned by a one-off $100 credit. The announcement came through the company’s Claude account on X and a corresponding support page update, closing a pricing episode that has been unusually public for a firm that normally ships without commentary.

The sequence is worth recalling, because it explains the awkwardness of the landing. Fable 5 arrived on 9 June as the first generally available model in Anthropic’s Mythos class, a tier sitting above Opus. Three days later a US export control directive forced its withdrawal for every customer, since the order barred access by any foreign national and the company had no way to verify nationality in real time. Access returned on 1 July after the Commerce Department lifted the order, accompanied by a promotional window scheduled to close on 7 July. It closed on 12 July, then on 19 July, each extension announced hours before the previous one lapsed.

Anthropic has been candid about the cause. Demand for Fable has been difficult to predict, the company said, which is why it staged the rollout and extended access as capacity was secured. That is a compute story more than a commercial one. Anthropic’s agreement to take the available capacity at the Colossus 1 data centre added more than 300 megawatts and in excess of 220,000 Nvidia GPUs to its fleet, and the effects have already shown up elsewhere: doubled five hour limits in Claude Code, the removal of peak hour throttling on Pro and Max accounts, and wider API rate limits.

The economics are unforgiving at the top of the range. Fable 5 is priced at $10 per million input tokens and $50 per million output tokens, the steepest published rate for any generally available Claude model. A single heavy agentic session producing 500,000 output tokens costs roughly $25 in credits, which sits on top of a subscription that may already cost $20 or $100 a month. For Pro subscribers, the $100 credit is a few weeks of serious use and no more.

There is a detail in the July 20 arrangement that has drawn comment from developers. The temporary 50 per cent uplift to Claude Code weekly rate limits expires on the same date, so the baseline against which the Fable allowance is measured shrinks at the moment the allowance becomes permanent. Max and Team Premium users therefore get half of a smaller number. Anthropic has said it continues to invest in capacity and expects the position to improve.

The competitive backdrop explains the reversal of the original plan, which was to remove Fable from subscriptions entirely. OpenAI released GPT-5.6 Sol to general availability on 9 July at $5 per million input tokens and $30 per million output tokens, half the input cost of Fable and three fifths of the output cost. On the Artificial Analysis Intelligence Index the two models score 59.9 and 58.9 respectively, close enough to be a statistical tie in general capability. Moonshot’s Kimi K3 landed on 16 July at roughly a third of Fable’s price. A $100 or $200 monthly plan that excludes the vendor’s best model becomes a difficult proposition when rivals include theirs.

What the Claude Fable 5 subscription arrangement reveals is a broader change in how frontier AI is sold. The flat rate subscription, which for three years implied access to whatever the vendor had built most recently, is separating into a bundled tier for everyday models and a metered layer for the expensive ones. Anthropic has drawn that line at the Max and Team Premium boundary. Others will draw it somewhere else.

For teams planning around this, three points are practical. Enable usage credits and set a monthly spending cap in the Console before 20 July, since access simply stops when a credit balance is exhausted and there is no automatic fallback to a cheaper model. Audit which workloads genuinely need Mythos class reasoning, because Sonnet 5 covers a great deal of routine agentic work at $2 per million input tokens. And treat any published Claude Fable 5 subscription terms as provisional until the billing page confirms them, a lesson this month has taught repeatedly.