maxcrit

1 million context window: Now generally available for Claude Opus 4.6 and Claude Sonnet 4.6.

Been trying to make Codex work for me but it overthinks everything, even simple stuff. Tried different reasoning levels, no difference. It just tries too hard. Good at reviewing others work though.

Giving sonnet 4 a 1m context window is kinda unhinged considering I see many folks struggle to keep it on task past 100k

STOP SAYING THANK YOU TO AI STOP SAYING THANK YOU TO AI STOP SAYING THANK YOU TO AI STOP SAYING THANK YOU TO AI STOP SAYING THANK YOU TO AI STOP SAYING THANK YOU TO AI STOP SAYING THANK YOU TO AI STOP SAYING THANK YOU TO AI STOP SAYING THANK YOU TO AI STOP SAYING THANK YOU TO AI

it's not even the model being "stupid". whoever is designing these integrations just... doesn't understand how to make the models do things for them or what makes llm agents work on a fundamental level. there's probably some overengineered scaffolding/single turn maxxed RAG topk vector search pipeline under the hood... ughhhh just thinking about it pisses me off

We have found one issue that leads to some users seeing inconsistent usage across sessions but it is quite rare, affecting less than 1% of users. We are working on a mitigation and continuing the investigation. For the rest we are not seeing evidence of higher usage consumption other than the advertised token cost increase of GPT-5.4 being 30% higher than GPT 5.2 and GPT-5.3-Codex.

Vibe coding is for experienced software devs. Do not vibe code until you really know how to do the basic stuff.