John Kler

Testing frontier AI models so you don’t have to.
Comparisons, use cases, and what actually matters.
17 | Learning in public on the AI frontier

[John Kler @JohnKlerAI]

@TokenGremlin Wow, so it sounds like OpenAI fixed all of their previous weaknesses. Or maybe they're not even fixed, but Astra is just so smart that it looks like it's fixed. Like @emollick 's "jagged frontier."

[John Kler @JohnKlerAI]

@NickPetros7 @ThePrimeagen Smart. Getting people to use the cheap models that make a ton of mistakes, and then selling the really smart and expensive models as the only fix.

[John Kler @JohnKlerAI]

@melvindvivas Did it feel sufficiently smart and fast?

[John Kler @JohnKlerAI]

@buildwithdylan Hopefully this is true.

[John Kler @JohnKlerAI]

@luckeyfaraday Have you tried Muse spark 1.3? Apparently it's not absurdly benchmaxxed and actually lives up to what the benchmarks say.

[John Kler @JohnKlerAI]

@TimJayas Oh my gosh, why is muse spark so good all of a sudden.

[John Kler @JohnKlerAI]

@Da7_Tech Are you serious? It's not absurdly benchmaxxed like I thought it would be? That's incredible if true.

[John Kler @JohnKlerAI]

@Da7_Tech Thank you for your service.

[John Kler @JohnKlerAI]

@kimmonismus I have a feeling we're going to see a big breakthrough before the end of the year.

[John Kler @JohnKlerAI]

@AndrewCurran_ What are the use cases for this extreme intelligence? Is Fable 5.1 not smart enough? Maybe I'm not being creative enough, but I genuinely can't think of anything below PHD-research level that Fable 5.1 isn't good enough for. I would rather have faster and cheaper models.

[John Kler @JohnKlerAI]

@pcshipp Cheaper and maybe quicker in the future.

[John Kler @JohnKlerAI]

@shiri_shh So excited for Grok 4.7.

[John Kler @JohnKlerAI]

@bridgebench Do you know how expensive it will be? Will it actually be usable on a day-to-day basis, or is it something I'll only be able to use a couple times a week before I run out of weekly usage?

[John Kler @JohnKlerAI]

@MyDaebakCafe @ashen_one Yes, but still technically benchmaxxed. A very impressive Flash model though.

[John Kler @JohnKlerAI]

@letsgetboosted @ashen_one Well, it is technically benchmaxxed because it obviously doesn't perform at Fable level, which is what the benchmarks claim. Still a very impressive release from the Google team.

[John Kler @JohnKlerAI]

@jonathan_wilke Have fun, just don't expect your usage to extend past a couple of days.

[John Kler @JohnKlerAI]

@beamnxw Is this not something that any AI automation workflow can do? What's so special?

[John Kler @JohnKlerAI]

@Yuchenj_UW Yes, Fable 5.1 talks and writes much better than Fable 5 or Opus. Looks like Anthropic listened to the people.

[John Kler @JohnKlerAI]

@ashen_one What defines "clipfarm stuff"?