Jan Piskac

AI-driven operations expert | Manga reader | Helping businesses streamline with tech | Love a good hike 🌄

[Jan Piskac @HoPik88]

@BoydBischke @trikcode Same experience. Everybody still talks about Fable 5 as a peak of AI, and it was when it first came out, but now it is nowhere near that level, and when he pleases it will switch to Opus 4.8.

[Jan Piskac @HoPik88]

@kimmonismus If it is GLM, it is worse than 5.3. When I run it side by side in opencode on backend tasks, it wasn’t near its capabilities, missed many nuances, and hasn’t gone deep enough.

[Jan Piskac @HoPik88]

@ArthurMacwaters Why do we have to look at a comparison where Opus and SOL use MAX thinking and Grok 4.6 only at high when we have the xHigh option?

[Jan Piskac @HoPik88]

@Im_IrushiK I know I mentioned only developer use😅
Gemini at its current state is completely out

[Jan Piskac @HoPik88]

@AstraiaAI Yes and it takes for ever to do anything. At token consumption through the roof.

[Jan Piskac @HoPik88]

@Da7_Tech And I thought it was just me. It’s taking forever to do anything, and the token consumption on max 20x it’s through the roof even though I’m only doing one project.

[Jan Piskac @HoPik88]

@blueemi99 Finally someone who sees it how it is. In my testing it is worse than GLM 5.3 and nowhere near SOL.

[Jan Piskac @HoPik88]

@Da7_Tech In my experience, you cannot trust just one model, no matter how good it is.

[Jan Piskac @HoPik88]

@morganlinton @VulcanBench I had the same experience, and Grok's build is truly amazing. They're putting a lot of love into it.

[Jan Piskac @HoPik88]

@BendikMatej Yeah, those were the good old times. Back then, even Sonnet was better than Opus when it hit newer versions than Opus, but that changed with Sonnet 5. Honestly, Sonnet 5 was a massive disappointment, and no one even knows what Haiku is anymore 😅

[Jan Piskac @HoPik88]

@thisisdimm And they should allow SuperGrok Plus users to use Team of Experts within the app as well. It consumes quota anyway, and we can use that in Grok Build anyway!

[Jan Piskac @HoPik88]

@BendikMatej Same, to be honest. I've been using Claude Code since it started and recently even cancelled the 20x MAX plan because I was cleaning the mess that Opus or Fable (if Fable didn’t decide it was risky and use Opus 4.8) did more than pushing forward.

[Jan Piskac @HoPik88]

@BendikMatej Well, each LLM provides different perspectives and views. I can’t just say GPT Sol is better in everything, but SOL is my go-to for deep understanding and planning. And when talking about Opus in its current state, I would prefer Grok 4.6.

[Jan Piskac @HoPik88]

@TimJayas It’s night and day, and not only that: even the way it responds in the Grok build is exactly what I want from an LLM, and in Cursor I couldn’t change that behaviour no matter what I tried.

[Jan Piskac @HoPik88]

@XFreeze I missed that by one day!! I was like, it’s 10pm, let’s do that tomorrow so I can have a full day, and they decided to drop the SuperGrok Plus 😅

[Jan Piskac @HoPik88]

@OmedVibeCodes Same experience and when I talk to them they just tried to hide behind "it is new frontier model and so on" and if you try to run research with swarm that's GG

[Jan Piskac @HoPik88]

@connerykirk @rubengarciajr In general, under Cursor you can get a lot of usage from Grok and Composer. I counted that on Cursor Ultra you would have around 2.5B tokens of Grok 4.6.

[Jan Piskac @HoPik88]

@Da7_Tech @Zai_org OK, I will pull the trigger and get the MaX for a year.