@FinanceYF5: 1/ Within hours of Fable 5's release, Twitter is in chaos. Karpathy says it's a major version leap. Some call it their "singularity moment." Others say they're starting to fear the future of software engineering. Today's top 10 reactions to watch.

1/ Within hours of Fable 5’s release, Twitter is in chaos.

Karpathy calls it a major version leap. Some people say it’s their “singularity moment.” Others say they’re starting to fear the future of software engineering.

The 10 reactions worth reading today:

2/ @karpathy’s review is the most direct.

Scores are SOTA across the board, but he emphasizes the feel:

Give it more ambitious tasks, and it just goes for it without hand-holding. The longer and harder the task, the bigger the gap.

“Never thought I’d directly skip reading code—but don’t do this in production.”

3/ Stripe’s 50 million line codebase migration.

A human team would need two full months.

Fable 5: one day.

Chubby says: the longer and more complex the task, the bigger the gap between it and other models. This isn’t just about benchmarks—it’s a full upgrade in efficiency, long tasks, and context management.

4/ Real-world system optimization: 2 hours beats everyone.

@VictorTaelin tests his own HVM5:

32 GPT-5 agents running for 20 hours → max 2x Opus 4.8 running for 8 hours → max 34% Fable 5 running for 2 hours → max 1770%

It also found a deep bug that the author himself missed.

“I don’t know what to say.”

5/ 3D forest scene, running in a browser.

@mattshumer_ says: Fable has solved 3D world building.

Built entirely with ThreeJS, fully custom, running in browser.

“Too absurd to speak.”

6/ Pokémon FireRed, pure vision, zero assistance, one-take clear.

Previous Claude needed map data, navigation aids, and game state info to barely run.

Fable 5 looks only at screen screenshots, thinks for itself, plays from start to finish.

No map reading, no code looking—just pure vision.

7/ One prompt: generate a stock market Minecraft roller coaster.

@JinjingLiang says: just gave one sentence: “build a minecraft style roller coaster for the stock market”

One-shot result. No iterative tuning.

8/ Space simulation with 5,000 celestial bodies.

Fable 5 vs Opus 4.8, same prompt, results side by side.

The gap needs no explanation.

9/ NYC skyline, four models compared side by side.

Fable 5, Opus 4.8, GPT-5.5 Standard, GPT-5.5 Pro, same prompt.

@aipulseda1ly says: “Best result I’ve seen so far, nothing more needs to be said.”

The gap is visible to the naked eye.

10/ Full realistic city storm scene.

Task: “Create a visually stunning shader that runs in the Twigl-dot-app environment. Imagine: an infinite city of neo-Gothic towers engulfed by a stormy sea with huge waves.” “Make the effect even better.”

11/ A discussion few noticed.

Fable 5 has a “silent limit”: when used to develop cutting-edge AI models, it won’t refuse, but it will quietly reduce capability.

Training pipelines, GPU clusters, model distillation—all within scope.

The model appears to help you, but it may already be holding back.

12/ Karpathy’s final words:

Software is becoming more like turning on a tap. Demand is actually expanding.

Have you tried it today? How does it feel compared to Opus?

That’s all.

If you like this thread:

1. Follow me (@FinanceYF5)
2. Like + repost the first tweet below

One sentence to make a game, one sentence to build a city, one sentence to recreate a classic.

Fable 5’s first day: absolutely insane!