Kaito

slow rider.

[Tyler @rezoundous]

ngl, seeing others saying GPT-5.6 is not as smart as Fable 5 made me a little sad.

[Kaito @kaitobythesea]

Amen to this

[John Ennis @johnennis]

This is yet another thing that really annoys me about Anthropic, claiming co-authorship of my work in git commit statements

Claude is not a co-author of my work any more than Apple is

It is a tool that I paid to use

5.6 Sol might be the end of me and Claude

[Kaito @kaitobythesea]

@theo Why do people feel the need to add the caveat that "it's definitely not Fable"?

[Kaito @kaitobythesea]

I mean it's anthropic, that's kinda what they do

[(((ل()(ل() 'yoav))))👾 @yoavgo]

j-lens is a nice technique and not just a renaming of hernandez et al 2023 -- taking the derivatives wrt to the future tokens and averaging is an effective idea. but the reporting over-reach like crazy re the implications.

[Kaito @kaitobythesea]

Man, I had real hope but the copium is starting to hit strong now

[Dan Shipper 📧 @danshipper]

GPT-5.6 is like a Porsche, Fable is like a warp drive.

We've been testing internally @every for about a month. And GPT-5.6 is the best combination of power, speed, and performance for your day to day knowledge work and coding.

Fable is a different beast. If you need to get across the galaxy use Fable. If you need to get around town using the best available tool for the job, use 5.6

Full vibe check dropping tomorrow!

[Kaito @kaitobythesea]

Descriptions of personalities here sound flipped from all previous models from these two. I'm a Codex guy but Fable has been my favorite model to date, and all the early previewers making sure to couch praise with "not as smart as Fable" isn't great.

[Mitchell Hashimoto @mitchellh]

I had early access to 5.6/Sol for ~month. Sol is my default. It is faster, plans/judges just as good as Fable, and I think produces better overall work. I’ll reach for Fable still for highly targeted debug or performance work with clear reward functions.

A cheeky way I describe Sol vs Fable to my friends is that Sol is a charismatic, efficient, talented coworker you’re jealous of. Fable is a genius recluse that is brilliant at its fixations but doesn’t go out, doesn’t date, and you don’t want to hang out with them much lol.

Fable is undefeated at highly targeted debug/security/performance goals. It’s a sight to behold and I was never able to get Sol to push as hard in this category. I’ll keep using it for this.

Sol is better or comparable at everything else, in my experience. Give it a shot, it’s hard to describe but it’s just more enjoyable to work with.

(Disclaimer I have no financial ties to either lab, wasn’t paid for any of this.)

[Kaito @kaitobythesea]

Based poaster

[Dan McAteer @daniel_mac8]

I have not been using GPT-5.6 for a few weeks.

I have no idea what it's like in practice or how it compares with Claude Fable.

Does that hurt my self-esteem as an AI poaster?

Yes.

Do I wish @OpenAI would invite me to be an early tester?

Yes.

[tender @tenderizzation]

data guys discussing their midtraining data mixes

[Kaito @kaitobythesea]

If you get Opus levels of usage at near Fable levels of performance, is that a worthy trade off?

[Dan McAteer @daniel_mac8]

Though I'm not sure...do we care about how GPT-5.6 compares to Opus?

Or do we only care about how it compares to Fable?

[Kaito @kaitobythesea]

My take is they had to put out some benchmarks to let people know they had a new SOTA model, but the other benchmark numbers weren't as clear cut. I'm guessing other benchmarks will be 50% win rate against Fable.

Still a solid release given the limitations Fable has.

[Dan McAteer @daniel_mac8]

GPT-5.6 Sol, Terra and Luna were announced by OpenAI and they only released a single popular industry benchmark.

It’s either:

A. GPT-5.6 crushes all benchmarks and they’re saving it for the release.

B. GPT-5.6 does poorly on other benchmarks so they don’t want to advertise it.

Which is it?

[Kaito @kaitobythesea]

@thsottiaux UI

[Yuchen Jin @Yuchenj_UW]

Anthropic, where is my Fable 5?

[Haiyu Wu @HaiyuWu1]

Working on world model or SSL? You definitely need to try our new work: VISReg!

What does it achieve?
💪 Strong collapse prevention: High gradient when embedding collapse
⚡ Friendly to scale training: Linear complexity to scaling factors
🧩 Easy to train: Similar to LeJEPA, it is a heuristic-free method
🏆 Best OOD performance: Achieving the best accuracy on 6 OOD datasets
📉 Data efficiency: Achieving a similar OOD average accuracy to DINOv2 with 90% less data
🧬 Robust to low-quality datasets: It is robust to long-tailed and sparse datasets

Our results also indicate that SIGReg type methods can scale up, filling in the missing piece in @ylecun's great talk https://t.co/P9TXmk3fFa.

A big thanks to my co-author @randall_balestr and my manager @DrMorganLevine. Also, huge gratitude to @ylecun for connecting us to make this project happen! 🤝

#SelfSupervisedLearning #JEPA #WorldModel

[Anthropic @AnthropicAI]

We’ve received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5.

We'll begin restoring access tomorrow, and will share an update soon.

We’re grateful to our users for their patience, and to everyone who worked with us on redeploying the models.

[Kaito @kaitobythesea]

The world isn't going to wait around. Delaying model releases just puts the US at a competitive disadvantage. The flywheel is turning, put the model in the bag and keep on trucking.

[Meituan LongCat @Meituan_LongCat]

Introducing LongCat-2.0 🐱
1.6T parameters · MoE with ~48B active · 1M context
The full model behind Owl Alpha on @OpenRouter — now available.

Built for agentic coding from the ground up:
◆ LongCat Sparse Attention (LSA) — scales efficiently for 1M-context tokens
◆ Zero-Compute Experts — dynamic activation 33B–56B per token, zero wasted compute
◆ MOPD — three specialized expert groups (Agent / Reasoning / Interaction), gate-routed per task

How it stacks up:
→ Terminal-Bench 2.1: 70.8
→ SWE-bench Pro: 59.5 (GPT-5.5: 58.6)
→ SWE-bench Multilingual: 77.3
→ FORTE: 73.2 · RWSearch: 78.8 · BrowseComp: 79.9

📖 Tech Blog: https://t.co/4KrjyKiDBn
Try it across different scenarios 🧵👇

[Kaito @kaitobythesea]

Bit the bullet and tried out Opencode Go today. First impressions of GLM 5.2 are super positive. Don't need much more than this to get normal work done. Best value in the game.

[Kaito @kaitobythesea]

This is actually nuts. Embedded intelligence in every page you visit. Opens up whole new avenues for builders.

[Xenova @xenovacom]

While we eagerly await Fable 5's return, our agentic WebGPU kernel optimization framework kept running.

Opus 4.8 picked up where Fable left off, pushing Liquid AI's new LFM2.5 230M to an unbelievable 1,400 tok/s... running locally in your browser.

Don't blink or you'll miss it.

[Kaito @kaitobythesea]

@thsottiaux Would be nice to have the agent get notifications from background processes like CC has so it doesn't have to spam sleep commands to monitor something long running.

[“paula” @paularambles]

okay midjourney, now make my spine more studio ghibli

[Kaito @kaitobythesea]

Good guy Tibo. 82-0. Never lost.

[Tibo @thsottiaux]

Dearest gentle codexer.

We did a sneaky double reset. Not only do you get a full reset on us. But you are also getting one into the reset bank to use at your own leisure.

Enjoy