cels

will stay in trenches until i make it

[cels @cels19x]

@uncledoomer if you use lovable genuinely what the fuck are you doing

[cels @cels19x]

@theo can you implement hermes agents into t3 code?

[cels @cels19x]

@maria_rcks Doing all this cuz of a clone is wild

[cels @cels19x]

@Tech_girl Ehh better but not more efficient

[cels @cels19x]

@bijanbowen How many params

[cels @cels19x]

Is it bad i still prefer fable 5?

[cels @cels19x]

@bridgebench I think fable 5 is way smarter and better to work with while opus 5 is the better “technical coder” type

[cels @cels19x]

@DavidOndrej1 lol

[cels @cels19x]

I think eventually Hermes agent will be the best coding agent officially. It without a doubt is the best “agentic” multi purpose harness. However as a raw coding harness i am bullish on Hermes. I still find Claude code to be the best coding but hopefully Hermes takes #1

[cels @cels19x]

@bridgemindai Who cares if it’s distilled. As long as it brings frontier intelligence cheaper I could care less

[cels @cels19x]

@bridgebench Do you even know what DeepSWE is lol it’s supposed to prevent benchmaxxing

[cels @cels19x]

@ChrisGPT prompt?

[cels @cels19x]

@TimJayas this is likely it however leaks seem to suggest they will be releasing a new gpt 5.6 checkpoint that will have improved efficiency.

[cels @cels19x]

@icanvardar fr get nearly the same if not better results with GPT. Opus is good at one shotting but not great with iterations

[cels @cels19x]

@thsottiaux YESSSS

[cels @cels19x]

@vaaselene Yeah its capable but super annoying to talk to and feels like it straight up doesnt understand intent. Ill be sticking to fable

[Mario Hachemer @MarioHachemer]

Genuinely and sincerely, f*** you Anthropic.

[cels @cels19x]

@sudoingX Agreed and followed

[cels @cels19x]

@Da7_Tech Average anthropic dick rider

[cels @cels19x]

@NovaXCode They sometimes just bake this into the training data