Codex lead accused of halving Pro limits as Astra trails Opus 5.5 on intelligence index
A long post breaks down OpenAI's cost-efficiency chart, noting Astra max scores 53 on the intelligence index, below Opus 5.5 high at 54, while costing 44% more.
>be tibo >"head of codex @ openai" >cut pro 200 limits in half before dev day >tell customers they'll "still get more work done" >with half the tokens lol >semianalysis drops a nuke proving the $200 claude plan gives nearly double the value of your $500 chatgpt pro >quick, deploy chart crimes to save face >"sol max $0.72 per task vs opus max $5.98" >"the real story is this one" >get caught immediately >sol 6.1 isn't even the same species as opus 5.5 >sol maxes out at 52 on the intelligence index >opus hits 51 on MEDIUM effort >fair comparison is at equal intelligence: >sol 6.1 max ($0.72) vs opus 5.5 medium ($1.34) >and opus medium burns fewer tokens per task anyway >oh wait, where is astra? your actual flagship model >the one @sama says is "AGI" >*checks* >astra max: 53 ($3.26) >opus 5.5 high: 54 ($1.82) >beats your flagship while costing 44% less >there is no openai model above 53 intelligence >it does not exist >everything above is anthropic only >opus 5.5 xhigh (56) >opus 5.5 max (58) #1 of 224 models >frontier means first >you know what the 2nd best is called? >absolutely nothing.
This post quotes another post: x.com/thsottiaux/status/2107295839655952486