/Tech1d ago

Claude AI releases Claude Fable 5, with tester Kieran Klaassen reporting performance that exceeds LFGBench limits

The Mythos-class model reportedly outperforms prior Claude AI releases

162K41135115.1K
Original post
Danielle Fong 🔆@DanielleFong#1002inTech

washed out relative to the quicktime capture :( quicktime doesn't work for this any more ?

grabbed a loop for my fellow VJs (video jockeys)

kudos to the beings that came up with like this 𝟝 symbolic imprint video technique

can you imagine if the interface worked like this? rolling across scaling and symbolic representations and meanings, broadband?

10:59 AM · Jun 9, 2026 · 407 Views
Sentiment

Positive users praise Claude Fable 5's benchmark-breaking capabilities while some criticize its Titanic remake demo as in poor taste.

Pos
66.7%
Neg
33.3%
3 comments with sentiment.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS62.4KBOOKMARKS58
Deedy@deedydas

Space simulations with >5000 objects

shirish@shiri_shh

Claude Fable 5 vs Opus 4.8

1dViews 62.4KLikes 136Bookmarks 58
LIKES1.9KRETWEETS39REPLIES12
lucy 🐧@uneventual

if you mention anything pictured in this animation to fable you will trigger the bio classifier and get sent back to 4.8

Claude@claudeai

Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use.

Its capabilities exceed those of any model we’ve ever made generally available.

1dViews 52KLikes 1.9KBookmarks 53
Kieran Klaassen@kieranklaassen

7/

real-time collab markdown editor, every span carries provenance (human vs AI), inline suggestions via margin cards, Yjs CRDTs over ActionCable, agent-native API — stack with almost no training data

Proof rewrite in Rails

https://pruf.kieranklaassen.com

1dViews 919Likes 3Bookmarks 8
Kieran Klaassen@kieranklaassen

2/

2-min animated short, 5 story beats, 3 TTS voices, 10 LLM-judge iteration rounds, everything procedural

Titanic remake in ThreeJS

https://emerald-glacier-58xy.here.now/

1dViews 960Likes 6Bookmarks 3
Kieran Klaassen@kieranklaassen

5/

auto-orbiting low-poly island, wind-sway trees, animated water, cozy structure, all geometry in code — no imported models

https://liminal-kernel-7yxm.here.now/

1dViews 371Likes 3Bookmarks 2
Kieran Klaassen@kieranklaassen

4/

isometric low-poly desert, click-to-walk, real breathing technique pacers (Wim Hof / coherence), plants grow as you complete sessions

Breathwork garden

https://centered-ponder-te8w.here.now/

1dViews 401Likes 2Bookmarks 2
Kieran Klaassen@kieranklaassen

3/

browser DAW, zero timing drift during playback, drum sequencer + piano roll, full mixer, no samples — synth only

Music production studio

https://wistful-shrine-939z.here.now/

1dViews 621Likes 2Bookmarks 1
Kieran Klaassen@kieranklaassen

6/

luxury rubber duck ecommerce — Next.js 14, live duck customizer, persistent cart, 3-step checkout with real form validation, statically exported

Rubber duck store

https://spicy-sparrow-mb3x.here.now/

1dViews 360Bookmarks 1
simon@disiok

@claudeai please buy more data centers asap

Jerry Liu@jerryjliu0

We've got 13 days to burn as much tokens as humanly possible on Claude Max plans

Before they revert to API based billing 💀

1dViews 3.1KLikes 6Bookmarks 0
Kieran Klaassen@kieranklaassen

fable-5 broke my benchmarks. here is some cool stuff I found.

Ran LFGBench across 12 prompts, here are a few:

Claude@claudeai

Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use.

Its capabilities exceed those of any model we’ve ever made generally available.

1dViews 10.2KLikes 43Bookmarks 33
Mike B@Mike_nike7

@kieranklaassen lol 'broke my benchmarks' is actually the best thing you can say about a model release. curious which prompt categories were the most different

1dViews 86
Bruno Bornsztein@brunotorious

@kieranklaassen Leonardo DiCaprio is turning over in his grave.

1dViews 25
blake@_buggles

@kieranklaassen That’s not the only prompt you’re giving it right?

1dViews 6
Emmanuel@emmanuelx0s

@kieranklaassen @danshipper Fantastic

1d