Composer 2.5 is presented as the leading coding model in an X post by @iruletheworldmo that weighs all factors and flags upcoming developments involving Elon

@iruletheworldmo If there's anything we can do better, or any weird quirks with the model, please let me know. We want to get the personality / behavior right!

Here I’d disagree. The model is still far behind GPT-5.5 and even Opus 4.7 in terms of quality. It’s not suitable for serious production use. It makes a lot of dumb mistakes and completely lacks the step-by-step reasoning that GPT has. Because of that, it produces poor results and then piles even more errors on top of them. This is yet another model with pretty benchmark charts, but for real-world work it’s complete slop

taking everything into account.

composer 2.5 is the best coding model in the world right now.

can’t wait to see what they do with elon.

@iruletheworldmo Bro what happened to GPT 5.6 yesterday?

@iruletheworldmo bro? I lost my life savings on polymarket

@iruletheworldmo Can’t stop praising it, so fucking good.

happy to see so many people being hyped about it

@leerob @iruletheworldmo the model truly surprised especially taking into consideration that i tried it out with a somewhat negative bias (i didn't like composer 2). besides all the great stuff the zig code it often writes is often a bit hilarious.

@iruletheworldmo Tell me you are not working on real software problems without telling me.

It is very good for the cost per token, but calling it the best in the world is a stretch.

@iruletheworldmo What's coming next will shock the Coding world, Elon is cooking

@iruletheworldmo Is it good at UI design? GPT-5.5 is quite poor at it.

@iruletheworldmo So bullish, this summer things are going to get totally wild.

@iruletheworldmo The interesting part is how fast “best coding model” keeps changing.

Feels less like a leaderboard now and more like whoever ships the tightest workflow around the model.

@iruletheworldmo Agreed! They somehow made a model that is as good on UI as Opus and as good on backend as GPT 5.5. And they made it fast and cheap. I´m not using anything else.

@leerob @iruletheworldmo I think for use in agent mode it’s a bit too ‘lazy’ out of the box - could be more proactive (instructions in agents md fix this but imo should be the other way around)

@iruletheworldmo is this true? what are people’s experiences with it, how good is it compared to gpt5.5?

@iruletheworldmo The moment they announced the collab with @xai and instantly got x10 compute than ordinary I said to myself mmm that’s another banger and here we go the results speak for itself

@iruletheworldmo Can't wait to see how _Grok_ is crushing coding

@leerob @iruletheworldmo Lee just continue whatever you are doing. You are a great Engineer and the product is the proof for it.

👊🏻🎯

@leerob @iruletheworldmo

@iruletheworldmo How much have you used it? It is still behind Opus 4.7 or even 4.6 MAX. There is something special about those models that 2.5 doesn't have