Dont you have gpt or opus?

both, gpt 5.4 includes codex 5.3.

Yes there have been reports of claude degradation.

I have gpt plus $20/mo but they seem to have really lowered the tokens for plus accounts.
I have claude Max 5x

Im sad nanogpt took GLM 5.1 and Qwen 3.6 out of subscription

I actually bought a service that Gave me 500M tokens (70$) forOpus 4.6/GPT 5.4 and Im about 2 weeks i used 360M tokens xD they Run out só fast

Trying to create a Dijkstra algorithm and background input for a bot for Maplestory

how was GLM 5.1 when you used it?

It was good, not as good as opus on its prime ofc, would maybe be similar now that opus hás downgraded

Qwen 3.6 surprised me most though almost 1 shotting A* algorithm

I am currently working on a FabricMC style modloader for a java mmo.
My prototype worked great, now working on the tooling so remapping all the ProGuard names from SLK4D23JF (class method field) names can be mapped to intermediary names with a tool to will reapply intermediary names based on bytecode analysis when the game updates and all the proguard names change. I mapped all the classes to human readable names already, got a couple 1000 method and fields renamed. This is going to be a huge token burner.

If you aren't setting your Effort level and/or have adaptive thinking enabled, then ya, you'll get mixed results.

Effort level is at maximum all the time

though GPT 5.4 is alot slower, it solved almost all problems in 3 small prompts

(everything else was basically opus 4.6 thinking on max effort)

When a new model comes out, its settings are always on high internally.. then after a week or 2 they see how much people are using it and how much resources its taking up.. to basically tune it to fit their current servers.. This is where it seems like they downgrade it. Personally IMO.. I think they also turn the current model down just before a new model is released.. so it makes the jump seem a lot bigger than it is.

Yea, that seems to be the trend