Anthropic just bought itself, and everyone still paying for Claude, one more week with its most capable model. On Sunday it pushed the deadline for free Fable 5 access from July 12 to July 19, the second time in two weeks the company has moved the date rather than let the meter start.
What actually changed
Since early July, anyone on a paid Claude plan (Pro, Max, Team, and premium Enterprise seats) has been able to run Fable 5, Anthropic's most capable and most expensive model, without paying anything beyond their subscription. That promotional window was set to close on July 12. It now runs through July 19.
Anthropic paired the extension with a second change that matters more day to day: it raised Claude Code's weekly rate limits by 50 percent. Fable 5 draws down your weekly allowance faster than any other Claude model, so heavier users tend to hit the ceiling well before any deadline. A higher limit means more real work fits inside the free window.
Nothing needs to be claimed or activated. The Free plan is not included, and API access is billed separately at the usual rates: $10 per million input tokens and $50 per million output, which makes Fable the priciest model on Anthropic's menu.
Why the date keeps moving
This is the second extension of a window that opened when Anthropic redeployed Fable 5 in early July. Each time, the company has said it will keep the window open as long as capacity allows and eventually fold Fable back into subscriptions as a standard model, though it has never attached a firm date.
The timing is hard to ignore. The latest extension landed the same week OpenAI shipped GPT-5.6, its new model family. Keeping its flagship free for a few more days is one way to keep people building their work around Fable right as a serious competitor arrives.
The part that actually changed my mind
What makes this more than a billing footnote is the head-to-head. I gave Fable 5 and OpenAI's new GPT-5.6 Sol the exact same builds: cloning a design app pixel for pixel, inventing a viral stunt, and building a learning platform from scratch. Sol came out surprisingly close. On some builds I could not tell which model did the better job, and Sol runs at roughly half the price, around $5 per million input tokens and $30 per million output.
That result is showing up everywhere this week. Developers are running Sol inside Claude Code through an official OpenAI plugin, wiring Fable and Sol into the same loop, and pointing Sol at long, multi-step jobs it can run for hours. The gap between the most expensive model and the merely very good one is thinner than it has been.
If you pay for Claude, the practical move is simple: get your Fable time in before July 19. As a pure orchestrator, the model that breaks a big job into pieces and directs the work, I still give Fable the edge. But that edge is thin, and it is worth knowing you have a cheaper option that is nearly as good.
How you use the week depends on what you already pay for. Without a Claude Code subscription, GPT-5.6 Sol on its own plans and builds the whole thing for about half the price. With both a Claude plan and a ChatGPT account, you can run them in tandem: let Fable orchestrate and hand the building to Sol and Opus underneath.
Either way, the countdown is real. Fable is included now, metered after July 19, and pay-as-you-go after that, until Anthropic decides to make it a standard part of the plan for good.