Kenny D

Builder. Stanford 2x. Dad 2x. San Francisco 🌉

[Kenny D @dkroshow]

Claude Code's combination of Fable 5.1 and Opus 5 is significantly better than Astra/Sol/Terra. Grok isn't part of the conversation yet but I hope it can get there.

Opus 5 somehow went from being the laughingstock of the AI world to a phenomenal worker. Also the mileage you can get against weekly limits is remarkable.

Astra simply can't compete with Fable 5.1 on being the trusted, frontier-level agent I'm looking for. Astra may get a better score on the test, but Fable 5.1 feels like a real person who has a good feel for the world. It also has a backbone, whereas I can easily convince Astra to go against something it just recommended. I also had this impression with Fable vs. Sol, but that calculus was different because Sol was much more token efficient. Astra destroys the weekly limit.

Anthropic is sort of the villain right now but you can't deny how good its product is

[Kenny D @dkroshow]

As a former Opus 5 hater, I've come to appreciate its capabilities, especially given it's basically free with Fable eating all of my Claude Code weekly usage

[Kenny D @dkroshow]

@JeffKazzee Agreed. Fable 5.1 royally botched a build for me recently

[Kenny D @dkroshow]

@alavery2 Love 1Password. Happy user for many years, and only getting better

[Kenny D @dkroshow]

Wild to me that terminal Codex and Claude Code still don't offer ways to filter out everything except agent and user messages

[Kenny D @dkroshow]

@morganlinton Not really an answer but this is my current view. I use a variety of benchmarks as input but then it’s all about vibes

[Kenny D @dkroshow]

Random thought on the frustration with coding benchmarks: the Michelin guide doesn’t over quantify a restaurant experience. Why do we do the same with these models?

Maybe the solution is to accept that just like so many things in this world, models can’t be reduced to pure numbers

[Kenny D @dkroshow]

Random thought on the frustration with coding benchmarks: the Michelin guide doesn’t over quantify a restaurant experience. Why do we do the same with these models?

Maybe the solution is to accept that just like so many things in this world, models can’t be reduced to pure numbers

[Kenny D @dkroshow]

@stefanoscalia Could not agree more

Camino Alto is an exception. They have a really good local rockfish dish. Arguably their signature entree at this point

[Kenny D @dkroshow]

Fable 5.1 and Astra are really hitting home the point that quality software development can't be fully outsourced to agents

[Kenny D @dkroshow]

@rish_gpt @bot @noahrshinn @alexandr_wang And the @LGAjots bot

[Kenny D @dkroshow]

@rish_gpt @bot @noahrshinn @alexandr_wang Bro your best bot here is the Paulie bot 😂

[Kenny D @dkroshow]

There's no way Fable 5.1 is cheaper than Fable 5

[Kenny D @dkroshow]

@quantumvoidlabs Amen!

[Kenny D @dkroshow]

I'd go as far as saying Astra is a regression for OpenAI.

I want an agent that I can trust with my highest level coding work. Fable 5.1 dominates in that regard. Astra can be unreliable, it often stops prematurely, and it destroys usage limits.

Fable 5.1 for planning and orchestration, and Sol/Terra for tier 2 tasks and below

[Kenny D @dkroshow]

@0xkvyb The problem is the subscription plans are so heavily subsidized the value is too good to pass up. Starting to fade away though

[Kenny D @dkroshow]

@moinulmoin @thsottiaux (@thsottiaux we legit appreciate you. Not trying to be critical here)

[Kenny D @dkroshow]

@moinulmoin @thsottiaux 😂

[Kenny D @dkroshow]

@woofadu I bet the pressure to stay at the frontier is so great that they had to ship before they could polish

[Kenny D @dkroshow]

100%, especially Astra -> Sol

[David Cramer @zeeg]

everyone using fable/astra (most of you on subs, because no one can afford that) should try switching back to the other high reasoning models (opus/sol)

you'll likely realize your tasks dont perform any differently