Composer 3 would be insane. Sometimes Composer 2.5 > Grok 4.6 Extra High?

Discussions
, ,
1

  1. I have been using Composer 2.5 OFF and ON since months. It’s insane. Fast AF. Nails UI. Not even frontier models enable fast iteration cycle as it. You never run out of tokens. No long 5 hour delays to continue! It is accurate, understands complex tasks and follows instructions with accuracy.

    Being Fast is what separates it from rest of the high cost claude/openAI Inaccessible expensive models where you pay to wait 2 weeks for your usage limit window to reset while you get overwhelmed by not being able to finish work.

  2. Grok 4.6 Extra High is not much better than Composer 2.5. Yeah it has high intelligence but at cost of time, TIME IS MONEEEY. Grok 4.6 consumes usage more and sometimes feel like Opus 4.6/7 with repetitive bugs it get stuck on, fails to fix. In short, wastes usage.
    Up till now, we could have gotten some Composer 2.6/7 based on Minimax newer models.

  3. I want Composer 3 to be a blend of Composer 2.5 speed and Astra level orchestrator and accuracy close to Fable. Might be a dream! But in future, we won’t even need prompting so…

    Composer 3 would be great if we had

    • Fast speed
    • Accuracy close to Opus 5/Astra
    • Good at verifying it’s work.

    Folks do try Composer 2.5, it’s really good at one shotting small applications. Composer has helped me with small tasks automated, cursor auto-triggers it. Composer 2.5 feels unlimited. Other models feel like slow doom into ruining your life and becoming a geek.

    Composer makes you feel like CHAD! Get your work done KING and GO enjoy your life.

Have a nice day, time to touch grass!

5

Yes please abandon Grok and rather advance Composer!

Grok 4.6 is so dumb at the moment!

8

Would love to see that as well! +1

9

Use it for small tasks!

10

I have no idea how anyone is getting superior results with Composer 2.5 over Grok 4.6. It’s not even close for me: Grok 4.6 spanks Composer. Maybe it’s just in my use case?

Having said that, Composer is based on Kimi (I think), so I assume Composer 3 will be based on Kimi 3, but that’s a beast of a model and is probably taking awhile to train.

11

for logic and even frontend work composer 2.5 performs well. and considering its price its impressive!

Grok is too wild and unpredictable.

12

That hasn’t been my experience at all, but there is so much variance in how people use this product, that I don’t doubt you at all. So much depends on the prompt (and built-in skills, rules, etc) that it’s no wonder users can have wildly different experiences with models.

13

Composer giving away much more reasoning in the output, more like chat gpt.
Plus for some projects composer could be used at 60-80% of simple tasks to cut the cost.

Something like auto-mod with Composer/Grok only would be awesome.
Manual switching between models that such an outdated approach.

So yeah, i hope composer 3.0 would be a thing anyways :slight_smile:

14

I’m so glad to know other developers are realizing the truth, acknowledging how good Composer is, instead of just blindly accepting Grok and following the hype around every new model release