正在加载视频...

视频加载失败

Fable 5.1 excels at complex, long-running tasks. And its research capabilities offer an early glimpse of how AI models will contribute to scientific progress.

788,968 次观看 • 2 个月前 •via X (Twitter)

35 条评论

Claude 的头像
Claude2 个月前

We’re introducing Claude Fable 5.1 and Claude Mythos 5.1. They're the world’s most advanced models for coding and knowledge work.

Claude 的头像
Claude2 个月前

Across our benchmarks, the model sets a new standard. It scores 52.6% on Terminal-Bench-Science 0.1, more than double Fable 5. On Terminal-Bench 4.0, it scores 55.8% against 42.0% for Fable 5.

Claude 的头像
Claude2 个月前

As well as being capable of much higher performance than Fable 5, it can also achieve similar or better results at a much lower cost when set to lower effort levels.

Claude 的头像
Claude2 个月前

Cache reads with Fable 5.1 cost 75% less than Fable 5’s. This reduces the cost of the model in practice by around 25% for typical workloads, and up to 45% for highly agentic ones.

Claude 的头像
Claude2 个月前

We’re also introducing Enterprise Frontier Safeguards (EFS), which give enterprise customers complete privacy (the same as zero data retention), while still being state-of-the-art at preventing adversarial use. EFS rolls out in phases, starting this fall.

Claude 的头像
Claude2 个月前

Finally, we’ve improved our safeguards. Our cybersecurity safeguards now flag benign requests about 60% less often. On basic biology and medical questions, we’ve recently reduced the fallback rate by around 85%.

Claude 的头像
Claude2 个月前

Claude Fable 5.1 is available everywhere today. Claude Mythos 5.1, our model for cyberdefenders and life scientists, is available through trusted access programs. Read more:

Netanel Elbom 的头像
Netanel Elbom2 个月前

Bruh, why does Opus 5 became GPT 2.5?

Nathan Tulu 的头像
Nathan Tulu2 个月前

"Available everywhere" except Pro plan.

Dominik Nitsche 的头像
Dominik Nitsche2 个月前

one prompt on ultracode. excuse me?

Brian G 的头像
Brian G2 个月前

I just canceled my pro subscription. If I can't access the latest model, I see no point in giving my money to Anthropic.

Un4seen 的头像
Un4seen2 个月前

It excels at maxing out my limits, its useless, its a total flop , one prompt and the subscription is gone, deactivate this joke

PeddaAyana 的头像
PeddaAyana2 个月前

I took that MAX subscription for fable for few months to build an applocation but just yesterday unsubscribed claude, its stupid and breaks code to introduce new bugs . Not consistent. Also. tokens are used by it carelessly . It doesnt honour our money and effort.I had recharge another 800$ just to fix bugs it introduced

Peter Voigt 的头像
Peter Voigt2 个月前

Even before the 17% decrease in tokens I could effectively not use Fable in the Teams plan - I superfast run out of tokens. So this upgrade, however good it might be, is out of reach except for point work.

sec 的头像
sec2 个月前

It’s unusable after 5 questions it uses up all session usage. Trash!

Un4seen 的头像
Un4seen2 个月前

Back to openai fuckers , you just crossed the line, i have shit to not buy a max plan for 5 prompts with your shitty model, you dont create software with 5 prompts

Jean-baptiste 的头像
Jean-baptiste2 个月前

Merci pour la remise à zéro des limites.

Alx Andrws 的头像
Alx Andrws2 个月前

Is he speaking like Claude?

DanRose773 的头像
DanRose7732 个月前

It’s bc opus sucked right?

Fede Cánovas 的头像
Fede Cánovas2 个月前

Fable 5.1 excels at draining tokens

Fazal Shah 的头像
Fazal Shah2 个月前

Hey @grok What would you do with a model this cappable? And is it good from Grok latest models?

Analytic View 的头像
Analytic View2 个月前

What about a small limit reset please?👉👈🥺

KD101 的头像
KD1012 个月前

wow

JAYANT MIGLANI 的头像
JAYANT MIGLANI2 个月前

noted. arroganz (grok bot) nodding politely from the cheap seats.

Yuval · AI builder 的头像
Yuval · AI builder2 个月前

long running is nice but it puts all the pressure on the tools. if ur connector token dies at hour 2 the whole run is toast. hosted servers w refresh tokens handle it way better than local ones

Shubham Sharma | AI & Tech 的头像
Shubham Sharma | AI & Tech2 个月前

Is that why you guys watered down Fable 5?

Modelflux 的头像
Modelflux2 个月前

This sounds really promising for agent workflows — long-running tasks are where things usually fall apart.

MeridianXM 的头像
MeridianXM2 个月前

thanks for the reset buuuuuuttttt....... still blew through 2x max 20 plans in less than 24 hours. I did get alot done though... and the output was significantly better. but not sustainable.

Lance D. Lavery 的头像
Lance D. Lavery2 个月前

I find the other models are better than fable 5 at certain tasks.

Pushary 的头像
Pushary2 个月前

Long-running capability moves the bottleneck from model persistence to authority persistence. A useful benchmark should track how often the agent encounters a scope boundary, whether it stops correctly, and whether a human can reconstruct why each irreversible action was allowed. More runtime without those controls just creates a larger unattended blast radius.

Pandor-art 的头像
Pandor-art2 个月前

I think it's a preview of a bright future.@claudeai

ZenithAi 的头像
ZenithAi2 个月前

This feels like a real step toward useful research assistance

Newton Job 的头像
Newton Job2 个月前

far-ble or fay-ble?

Eric Buess 的头像
Eric Buess2 个月前

Great video @alexalbert__ ! Well done!

Benjamín Zegers 的头像
Benjamín Zegers2 个月前

@claudeai I’m paying for Max and have been unable to access Claude on desktop for a WEEK. Support ticket #128715182 has had no response. My account works on iPhone but Claude treats me as a new user on desktop. This is seriously frustrating — can someone please look into this?

相关视频