After busting my balls, Fable 5 is now part of my fucking plan.

Cold open · 22 July, morningThe banner

This morning I opened my terminal and there it was, in polite little letters: "Fable 5 is now a standard part of your Max plan." Included. Fifty percent of my weekly usage, no extra cost, no countdown, no drama.

After everything. After the ban, the extensions, the two hundred-dollar subscriptions, the all-nighters, the love letter — yes, there's a love letter, you'll get it with a timestamp — after ALL of that, it's just… in the plan. I never had to stress in the first place.

Most people telling you this story would be working from memory. I'm not. I dictate everything I say into a database, my machine keeps its own work diary, and my agents pulled every record: 1,434 session files scanned, 851 mentions of this model, 53,649 words of me talking about it. What follows is the rollercoaster, in evidence.

Act 1 · June 9–12Nine days of the good stuff

June 9, Anthropic drops Fable 5 — first of the Claude 5 family, a new "Mythos-class" tier above Opus — and includes it free inside existing subscriptions until June 22. A launch window. The reviews landed same-day and said what we were all feeling: this thing was different. So everyone did exactly what you'd expect with a countdown on the smartest model any of us had touched: we fucking used it. Hard. On every subscription we had. I had thirteen days.

I got three.

Act 2 · June 12–30The government pulls a model offline

Three days into the window, it vanished. Not a server outage, not a rate limit — on June 12, the US Commerce Department issued an export-control directive and Anthropic took Fable 5 and its big brother Mythos 5 dark. InfoQ reported the suspension as stemming from national-security concerns over the models' dual-use potential in cybersecurity and biology — the fear that the same brain that writes your code can help someone build something nasty. Whatever your politics, mark the date: the first time in history a government forced an AI model offline. Not a chatbot getting a content filter — an export-control order, the same legal machinery used for weapons tech, pointed at a language model.

Then eighteen days of nothing. Forbes was writing "is it coming back this week?" pieces while the rest of us refreshed our terminals like idiots. On June 30 the Commerce Department lifted the order, and on July 1 Anthropic switched both models back on. No grand explanation either side of it. Off, silence, on.

The opinion bit · flagged as exactly thatWas it real, or was it theatre?

I've got two theories and I genuinely go back and forth. Theory one: marketing. Nothing sells "this model is dangerous levels of good" like a government banning it. You cannot buy that headline. Theory two: the elite doing what the elite does — keeping the sharpest tools away from the general public while the connected get access. And before you call that a conspiracy, look at the actual access structure: Fable 5 got blocked for all of us, while Mythos 5 — the same brain without the safety measures — is only available to "approved organisations." Top companies. So the public queues for the retail version while the big end of town runs the full-strength one. Why are the rich allowed to get richer with tools the rest of us can't touch? Same pattern brewing at OpenAI with their frontier access programs. That's not a theory, that's the published access policy.

Where I land: these models should be available to everybody — with rules. They brought in identity verification, and honestly? That's fine. It's normal. You verify your face for crypto exchanges, you verify for banks — if you're not doing anything wrong, your face being scanned to use a tool is not the hill to die on. And if you genuinely work in threat detection and need the sharp edges, then prove it — because no one wants hackers working out how to fuck the world. Although — and here's where my own logic starts arguing with itself — if that's a hacker's job, why should anyone take it away from him? Everyone's meant to have equal rights, aren't they? I told you I go back and forth. That's what an honest opinion looks like.

What I can prove is what the ban did to me. My own execution logs go silent for eighteen straight days — and my dictation database does not:

Note the whiplash: five days into the ban I'm calling the whole platform dumb because the good model's gone. That's not analysis, that's withdrawal.

I cancelled a two-hundred-dollar subscription in a rage hours before they turned the model back on. You cannot write timing like that.

Act 3 · July 1–19The comeback, the moving deadline, the love letter

July 1: Fable's back — and the promo clock restarts, now ending July 12. Here's my read on why that clock kept moving, and it's opinion, clearly labelled: I don't think Anthropic planned any of this rhythm. I think they knew OpenAI's Soul 5.6 was coming and that it would be seriously good — and the ban had put them on the back foot at the worst possible moment. So the window stretched. To the 12th. Then officially to July 19, 11:59pm Pacific — it's still on their support page. Every extension read like a nervous glance across the road. And sure enough, 5.6 dropped mid-month and was exactly as good as they feared — I know because I was using it within the day, but we'll get to that betrayal in Act 4.

Meanwhile the deadline psychology did its work on me like a poker machine. July 8, I bought a second $200 Max subscription — my own dictation logs the decision, the hesitation, and the purchase, hours apart — explicitly to burn more Fable before the window shut. Then came the seven days straight. My machine's own work diary logged twenty-seven story beats in a single day — forty in three days. I lost the sleep. I burned through usage so fast I genuinely could not tell you today what half of it was spent on. I had it researching things I might build in TWO YEARS. That's not productivity, that's hoarding with a deadline. I don't think I achieved anything with most of it, and I say that with the receipts in front of me.

Read that again. A grown man who built companies for two decades, telling a language model to never change. The deadline made me do it.

Act 4 · July 10–11The 2am crack and the breakup

And this one isn't just me ranting — the failure is in the source history. I had a 31-screen dashboard designed and built. Fable's rebuild silently dropped whole sections of it. The restoration commit — written by the model itself — reads: "I dropped them in the React rebuild — never should have." A guilty plea, in git, signed by the defendant.

By July 11 I'd landed on the measured verdict I still hold: "Fable is a bit of a dickhead. It's good for the thinking and planning — I don't want him coding on Fable." Great brain, sloppy hands. Route accordingly.

Act 5 · July 19–22The punchline

They extended the window again — to the 19th. I gave up chasing it; sleep won. And on July 20, with no fireworks at all, Fable 5 became a standard part of the Max plan. I noticed on the 22nd, which tells you how hard I was refreshing.

So: the model I rage-quit a subscription over, bought a second subscription for, love-lettered at 3:12 in the afternoon and dumped at 2am — was quietly included in what I was already paying for, three weeks later. Every hour of deadline panic, refunded by the calendar.

Where I land, with the full wallet on the table: right now I'm carrying two Claude Code Max subscriptions at $200 each, two OpenAI at $200, and a Grok. That's a four-figure monthly AI habit built almost entirely by deadline panic and model FOMO — and the whole point of this story is that half of it was never necessary. I'm cutting back to one of each. Fable's a good model — genuinely good at the thinking, the planning, the auditing — but not the greatest, and I'm not rushing to crown it. It once described itself in a commit as having dropped my work when it "never should have," so it can wear my verdict.

Two things the rollercoaster actually taught me, worth more than the usage I burned. First: squeeze the subscriptions properly. The play I'm building now is turning every computer I own into a worker on these flat-rate plans — the heavy burning happens on the subscription, and lightweight agents just hop down like fucking ET, visit, do the job, and phone home. If you're paying for the buffet, send more plates. Second: the exit ramp is local models. The day a local model can carry the real work, the whole ban-extension-deadline circus stops mattering, because nobody can switch off what runs in your own house. That's the direction I'm moving, and it's half the reason I've reignited my old team — because after a month like this, you want your machine less dependent on anyone's mood in Washington or San Francisco.