KI WOCHE

Die deutschsprachige Plattform rund um KI -

[KI WOCHE @kiwoche]

interesting

[BridgeMind @bridgemindai]

Fable 5 CANNOT be removed from Anthropic subscriptions.

The new Artificial Analysis Coding Index shows exactly why.

Without Fable 5, Anthropic's best subscription model is Opus 4.8. It loses to GPT 5.6.

It even loses to GPT 5.5. The OLD GPT.

At that point there is no reason to pay for Claude. None.

Fable 5 is not a perk.

It is the entire case for the subscription.

[Magnus @getmagnus]

Day 3 of using Fable 5 & @threejs to make a horse game.

Lots updated.

[Build Fast with AI @BuildFastWithAI]

We tested Fable 5, Grok 4.5, GPT 5.6 Sol Ultra, and GLM 5.2 on the same arcade game prompt.
This was the result :

Claude Fable 5 - the smoothest gameplay of the four. fluid movement, no jank.

Grok 4.5 - nailed the screen shake, particle explosions, neon cinematics. most visually dramatic build.

GPT 5.6 Sol - the all-rounder. genuinely great gameplay experience, nothing broken, everything solid.

GLM 5.2 - an open-weight model shipping a complete, playable build that held its own against three closed frontier models.

Which one did you like the most?

[KI WOCHE @kiwoche]

wtf

[Chris @ChrissGPT]

5.6 Sol and what I assume is - 750 TPS in the wild 👀

This video is not sped up.

[Aniket J @aniketjart]

Grok 4.5 is insane for game dev
Built with Grok 4.5, @usecrayon on #threejs

#gamedev #grok

[ClaudeDevs @ClaudeDevs]

We've reset 5-hour and weekly rate limits for all users.

[KI WOCHE @kiwoche]

Anthropic has reset weekly limits.
So happy GPT-5.6 release day with Fable 5, I guess?!
#Fable #Fable5

[🚨 AI News | TestingCatalog @testingcatalog]

ANTHROPIC 🔥: Claude now has a Reflect feature, which produces a monthly summary of your interactions with Claude.

> The feature is now available in settings for Free, Pro, and Max users in Beta.

> Fun fact: During development, this feature was named Cardinal.

* It seems like it is rolling out slowly and may take some time until it arrives.

[KI WOCHE @kiwoche]

#Opus5?

[Chubby♨️ @kimmonismus]

Looks like Opus 5 is on its way.

That was to be expected and foreseen, given that Fable 5 and Sonnet 5 have already been released and only Opus 5 is still missing.

Especially since Fable 5 is being removed from the subscription plans at the end of the week, the "permanent underclass" needs an affordable model - one that is good, but not *too* good - to keep them within the plan.

[BijanBowen @bijanbowen]

the X feed today be like...

[KI WOCHE @kiwoche]

true story

[Vivek Sen @Vivek4real_]

"YOUR CLAUDE CODE SESSION LIMIT HAS BEEN REACHED"

[KI WOCHE @kiwoche]

Focus. You are overdoing it with too many overlapping tools.
The Google AI product portfolio in many ways feels like an experimental playground. So the adults who know what they want go to Claude and Codex.
Don't mess up great technology with a confusing product portfolio and lack of positioning.

[KI WOCHE @kiwoche]

@AiBreakfast No. Someone finally needs to reset the weekly limit. 🥹

[KI WOCHE @kiwoche]

🤓

[Tech Dev Notes @techdevnotes]

Grok 4.5 is #1 on Harvey's Legal Agent Benchmark

[KI WOCHE @kiwoche]

🫣

[Husk @huskirl]

Had to give the new voice model the classic test

[QASIM-livelifewithai @ggg78g89]

SWE-1.7 looks like a serious leap for coding agents.

Frontier-level performance at lower cost, with 1000 tok/s speed, is exactly where this race gets dangerous.

RL is clearly not done scaling, and Devin is pushing the merge-ready coding frontier hard. This is huge for builders.

[Cursor @cursor_ai]

We've partnered with SpaceXAI to train Grok 4.5.

It’s our most powerful model yet and the first we've built for more than software engineering.

[SpaceXAI @SpaceXAI]

Announcing Grok 4.5, our first model trained specifically for coding and agents. It was trained with Cursor and offers frontier intelligence at leading speeds and cost efficiency.

https://t.co/i8HpU7w64k

[KI WOCHE @kiwoche]

@nichochar Why did you close it?

[LTX.io @ltx_io]

Read the full announcement → https://t.co/jaTdJZSQoK