GPT-6 Sol

“i’ve noticed this since GPT-6 SOL and honestly I do not mind the work taking longer for it to complete as long as I get more usage out of it.”

“Feels not as good as 6sol”

“Generated code (I mostly program C++) is often bloated and not succinct or "simple" enough to my tastes”

“took 80% of the pro plan in 3 tasks”

“Now this feels like its the actual Sol 6 that was meant to be.”

“GPT-6 Sol was released on September 22, GPT-6.1 Sol was released on September 29, with just one week in between”

“I used 6SOL to develop something for 2 days, but because it was too bad, I had Astra rewrite it”

“I dont trust code reviews by it”

“GPT 6 Sol still beat Opus 5.5 on code review”

“After a banked reset your weekly usage gets cut by about 80%”

“6-Sol is noticeably worse for code reviews”

“twice as much usage on Sol (which is equivalent to what you use to get on Sol anyways”

“GPT 6 Sol has been borderline unusable today for me, extremely slow, doesn't follow instructions, and is writing very stupid code. I tried 5.6 Sol as well and it seemed to have the same issue.”

“It completely erases my anxiety about the usage limit”

“6 can at least get through to the end”

“sol made some really brain-dead decisions - even though I had a pretty detailed plan”

“Put the same bug into Codex to fix, DeepSeek 21 seconds, Sol 58 seconds, both fixed it correctly”

“GPT-6 Sol's max effort score is 68.8%, only 1.1 percentage points behind Claude Fable 5's top score of 69.9%, but the per-task cost is about 80% lower”

“turns into poke-and-it-moves”

“ignore your instructions in implementation”

“6 Sol did a better job reviewing and configuring my AWS setup than 5.6 did”

“In the official charts, GPT-6 Sol runs DeepSWE at $2.74 per task, the previous generation was $6.46”

“I made GPT 6 Sol orchestrate a GPT 6 Luna to implement something on a music player I'm working on, it took around 20 minutes and spent only 1% of my weekly usage.”

“After using it, the tuning feels quite conservative; it has to ask for confirmation on everything. It should be a strictly controlled version.”

“I can get 72hours or more of usage running both at the same time”

“GPT 6 sol, while it seems to be controversial, can go on for seemingly forever”

“on med or even high it makes errors with xhigh i find it surprisingly capable of self correcting”

“no safeguards observed”