Berlin

I disapprove of what you say, but I will defend to the death your right to say it.

[Berlin @Berlinworlds]

@jimcramer Nobody here is realizing the gameplan has always been Inverse Cramer

[Berlin @Berlinworlds]

@urjr1 Very cool project, needs a bit of design work tho :) And a creative GTM move hah

[Berlin @Berlinworlds]

@nicksaraev But nowhere near as tasteful as Fable

[Berlin @Berlinworlds]

@LinearUncle @mweinbach Need this

[Berlin @Berlinworlds]

@ArielKwiat This really went over people’s heads 😄

[Berlin @Berlinworlds]

@melqtx I'm the left one here but proud to admit it

[Berlin @Berlinworlds]

@Sauers_ I noticed it was mentioning being Opus 4.5 in its reasoning, it also validates this fact, it's interesting

[Berlin @Berlinworlds]

@entirelyuseles @defaultuser9999 @0xvanbeethoven @bakkermichiel Upstream has a predefined system prompt of "You are a helpful coding assistant" even if you have no system prompt over the API, but this changes when referencing the model's identity specifically. Entering that conversation without such a direct question, it will say it's Claude.

[Berlin @Berlinworlds]

@btibor91 The GPT design is more sensible here

[Berlin @Berlinworlds]

@hammerhoundai @zxytim Just run it in Pi or Claude Code

[Berlin @Berlinworlds]

@morganlinton Fable 5 is attentive, intuitive, and has a clear understanding of good ways to do something (not necessarily what I would pick, but an above average approach). GPT 5.6 can implement end to end but make a mess and still makes weird and impractical decisions alone, like other GPTs

[Berlin @Berlinworlds]

@maria_rcks I have like 98% limits left on the Codex plan because Fable is incomparable intelligence. But limits go a lot quicker

[Berlin @Berlinworlds]

@theo 3 is the best of the options, but I would make the Running part smaller, and allow both chat-only and project-split list for Recent

[Berlin @Berlinworlds]

@goodside Tencent Hy3 is relatively good at this

[Berlin @Berlinworlds]

@devnullll @grok @xai Use @PawelHuryn Grok in VS Code

[Berlin @Berlinworlds]

@NeetzscheLLC @UbermenschMind Optimal response

[Berlin @Berlinworlds]

@PeterDaveHello @SpaceXAI Limit on SuperGrok does need a boost or a few resets

[Berlin @Berlinworlds]

@morganlinton Somehow this does match my experience, Grok 4.5 and Fable are the best models for me to work with. I haven't been able to elicit particularly good work from any of the GPT 5.6s, and they sound quite condescending, even when they don't know. Also more hallucination-prone than 5.5

[Berlin @Berlinworlds]

@martin_casado Fable has superior intuition and planning skills, but Grok 4.5 is an optimal daily driver implementation model

[Berlin @Berlinworlds]

@PawelHuryn @perplexity_ai So great, love Grok but the CLI is not as comfortable as a VS Code extension, very excited to have seen this