Ari

Building stuff. Feedback always welcome.

[Ari @arimxyer]

@thsottiaux I tried this. And it's genuinely insane how well it runs :o

[Ari @arimxyer]

@gary_miklos @levelsio He uses commits the same way everyone should largely use commits. As a “save” button. More VCS, Less SCM.

[Ari @arimxyer]

@ctatedev

[Ari @arimxyer]

They can't pick a lane because Fable 5 doesn't fit either one.

Retention isn't a setting on this model — it's architecture. Fable 5's safety classifiers require prompts and outputs to be retained. Which means it structurally cannot be sold under Zero Data Retention, the thing Anthropic's enterprise book is built on. And 30 days is the floor, not the ceiling: flagged content can be held up to two years, and that clause can't be waived under an existing ZDR agreement.

The receipt: Microsoft pulled Fable 5 from the model picker its own employees use in internal GitHub Copilot — while selling it to Copilot and Foundry customers the same week. Opus 4.8, Sonnet 4.6, Haiku 4.5 all stayed. They're still under ZDR. Redmond's lawyers read the terms and said no, and Microsoft has every commercial reason to say yes.

So "put it under the API already" isn't actually available to them. The buyers who'd pay for it can't accept the terms.

Which leaves subscriptions — at $10/$50 per Mtok, double Opus 4.8, the priciest model they've ever listed. That doesn't survive contact with an agentic Claude Code loop. Anthropic says the constraint is capacity. Fine. Both things can be true.

So you get this instead: five access decisions in 33 days. Capped launch, export-control blackout, meter announced, extension, extension. Every reset teaches people to burn frontier tokens on work that doesn't need them, because the window might close. Anthropic controls the tap AND the countdown. Intentional or not, that IS a pricing strategy.

It's not games. It's a model with nowhere to go — too retention-hostile for enterprise, too expensive to bundle — and a countdown doing the work a decision should be doing.

[Ari @arimxyer]

HAHAHA @OpenAI really f*cked em up

[Claude @claudeai]

We're extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19.

[Ari @arimxyer]

@samhogan Wowowowowow- this is legit

[Ari @arimxyer]

“Mom… They’re fighting again…”

[Ari @arimxyer]

The only benchmark standard that matters

[Diego | AI 🚀 - e/acc @diegocabezas01]

Same prompt 7 months later

[Ari @arimxyer]

@levelsio This is a meme right? This can’t be real

[alex duffy @alxai_]

Don't agree with much of AI 2040.

Do think we really need more proposing & discussing of visions for the future.

So turned theirs → a web-comic,
to broaden that discussion.

[Ari @arimxyer]

Genuinely feels like a “ChatGPT” moment.

[Ari @arimxyer]

@OpenAI’s best product, following the recent releases, might not even be the 5.6 family (although stellar). I think what will be remembered for the long haul from this batch is GPT-Live.

[Ari @arimxyer]

@gmi_cloud Have you tried both on max? And if you ran 5.6 again, same effort, second prompt. Would it improve the game further for still less money than fable 5?

[Corey Quinn @QuinnyPig]

I'm really confused by @OpenAI's strategy for GPT-5.6. I can't find the date the pricing jumps by 50%, the date it leaves "being included in subscriptions" is likewise missing, and what their employees say here aligns with what the documentation and blog posts say.

[Ari @arimxyer]

@MarcosHernanz What provider are you using?

[Ari @arimxyer]

@ClaudeDevs GOATs

[Ari @arimxyer]

@DanielLockyer https://t.co/Sws9foT39X

[ClaudeDevs @ClaudeDevs]

https://t.co/dxoI9Nna86

[Ari @arimxyer]

@ctatedev f*cking sheesh dude. And I was just getting used to dioxus.

[Ari @arimxyer]

😁

[dax @thdxr]

i've never hyped a model release, we're generally conservative with how we use these things

but gpt-5.6 has had a massive impact on our team, we're using 5x the tokens as we used to

it's not even smarter than fable or anything, but it's just so reliable and fun to use

[Ari @arimxyer]

@benhylak So dumb.