MaxMusterman

[MaxMusterman @maxschuetz_]

What will happen if we get opus 4.8 quality in a open source model in a year??? What's even the limit of Frontier LLMs?

[MaxMusterman @maxschuetz_]

@Yiliangs1 @cjzafir codex is better and espacially in reviews it catches so many edge cases. Claude has actually implemented an own rule: Codex always reviews.....

[MaxMusterman @maxschuetz_]

@ClaudeDevs i cant change the effort any clue why?

[MaxMusterman @maxschuetz_]

@ClaudeDevs Holy shit

[MaxMusterman @maxschuetz_]

@edgarpavlovsky it kinda depends, what kind of work. I am currently fully automating my bug/issue work. No prompting to fix a bug that a user has reported, since the user that reported it, basically prompted enough.....

USER DATA TO PROMPT YOUR CLAUDE

[MaxMusterman @maxschuetz_]

claude code limits are somehow much better..... i am not really running low on my limit. In 8h it is resetting and i have still 8% left.

[MaxMusterman @maxschuetz_]

@Samaytwt i dont what you are doing... but i am running multiple agents, with quite a lot of overhead (memories etc.) and i dont geht to my session limits ever.

[MaxMusterman @maxschuetz_]

@petergyang Symlinks are my anwser

[MaxMusterman @maxschuetz_]

codex 5h limit sucks...

[MaxMusterman @maxschuetz_]

@DanielSmidstrup claude for ideas, spec & planning. Codex on backend and cross review. Finds some diabolical stuff. Claude always asks codex to review and made that his first rule.

[MaxMusterman @maxschuetz_]

codex limit reset

[MaxMusterman @maxschuetz_]

@EchoTruthHub @ClaudeDevs it works pretty good for me to embed it into the skills I use. For example /autopilot skill which has a hard rule, that the orchestrator should make sure, that everything has been reviewed enough. Almost every task needs one review round to fix stuff...

[MaxMusterman @maxschuetz_]

@AnthropicAI I need a reset please....

[MaxMusterman @maxschuetz_]

@AGIGuardian @ClaudeDevs Use subagents and codex reviews.

[MaxMusterman @maxschuetz_]

@EchoTruthHub @ClaudeDevs use subagents and cross model review via codex

[MaxMusterman @maxschuetz_]

with all the claude opus 4.8 bashing, its good to use the claude IDE. Shared View across 6 Sessions is actually amazing. You can have one screen determined to all your sessions and you are good. Nice work.

[MaxMusterman @maxschuetz_]

"Notable: Codex (Stage 3) caught 3 real bugs both Claude reviewers cleared" - Claude Opus 4.8 extra

[MaxMusterman @maxschuetz_]

Claude Opus 4.8 xhigh is sometimes admitting that gpt 5.5 is essential.....

[MaxMusterman @maxschuetz_]

i have used claude code ide for the last week or so. I really like it and I am positively surprised! Opus 4.8 is also now running more like GPT 5.5 in Inline Sessions.

[MaxMusterman @maxschuetz_]

workflow feature is cool but expensive. I just burned 4-5% of my weekly limit...