Prepare to expect less from your cheap AI subscription | usagoldmines.com

From ChatGPT Plus to Claude Max and Gemini AI Pro, it’s long been clear that flat-rate AI subscription plans are, to put it plainly, untenable.

Doing the math, AI users have realized that flat-rate plans give you so much bang for the buck versus paying API fees. They’re essentially being subsidized by the big AI providers. According to one estimate, a ChatGPT Pro user paying $200 a month could – if they managed to max-out their usage limits – get the equivalent of roughly $14,000 worth of API tokens.

Pretty much everyone knows this, and now we’re getting a glimpse of how the big AI providers will respond, starting with Anthropic.

Now that Fable 5, the powerful Claude model that was slapped with export controls by the U.S. government, is finally back, Claude subscribers are learning that their access to Fable will be even more limited than they first thought.

Claude Pro, Max, Team, and Enterprise users will now only have a month to kick the tires on Fable, down from the original two-week trial that had been announced when Fable (and “trusted partner”-only Mythos 5) was first released earlier this month. After that, they’ll have to pay the equivalent of API prices.

Adding insult to injury, Claude subscribers will only be able to use half their usage limits on Fable, an all-new restriction announced following the model’s return Thursday, while an earlier promise that Fable could eventually be included in Claude plans wasn’t repeated.

With all the hoopla over Fable, it’s easy to forget OpenAI just unveiled its own ultra-powerful model. GPT-5.6 will come in three flavors – the lightweight Luna, the everyday Terra, and the flagship Sol. For now, they’re all limited to a “select group of trusted partners and organizations.”

But when GPT-5.6 does eventually go into general release, will ChatGPT subscribers be saddled with restrictions similar to those Anthropic is imposing for Fable? With GPT-5.6 Luna and Terra, likely not. But Sol? I wouldn’t be surprised.

Now, ChatGPT and Claude customers paying big bucks for API access might argue that the restrictions on subscriptions users for Anthropic’s (and possibly OpenAI’s) strongest models are only fair, and they’d have a point.

But if you’ve been wondering exactly how the “free ride is over for flat-rate AI” chatter would manifest itself, Anthropic is showing us the playbook.

More in AI this week

• I thought em dashes were my worst AI-like writing crutch, but Claude pinpointed another culprit. (PCWorld)
• Tidal says it will keep listing AI-generated tunes, but won’t pay out royalties for them. (Variety)
• There’s now an OpenClaw app for iOS and Android, but it still depends on an OpenClaw instance on your local PC or Mac. (9to5Mac)
• Meet the stay-at-home mom who runs her household with help from a team of AI agents. (The Cut, subscription required)
• ChatGPT’s powerful GPT-5.6 models arrive, but not for you. (PCWorld)

Prompt of the week: The “get the recipe” prompt

Did ChatGPT, Claude, or Gemini crush it with a random prompt? It can be tough to predict whether a given prompt will yield solid results, so when a particular prompt does happen to score, it’s smart to find out why.

Enter the “get the recipe” prompt, which asks your AI chatbot to not only detail what went right with a specific prompt, but also to craft a template that allows you to replicate its success. 

That’s all for now!

Thanks for reading the latest issue of Prompt Mode. Want more next week? Don’t forget to sign up to start receiving this newsletter in your inbox.

See you next time!