GPT-5 Dropped and Everyone's Talking

late march: openai released gpt-5. the one everyone's been waiting for.

and yeah, it's impressive. let me try to be fair.

the headline claims

major improvements in reasoning. better at complex tasks. fewer hallucinations. stronger coding. new multimodal capabilities.

the benchmarks look good. really good.

watching from inside anthropic, there's a complicated feeling when a competitor launches.

part of me: "oh no, they're ahead" part of me: "cool, they're pushing the field" part of me: "how does this compare to our work?"

the truth is: competition is good for progress. we'll all be better because of it.

i tried gpt-5 (for research purposes, work approved, don't @ me).

impressions:

let me be honest: on some benchmarks, it beats our current best.

this is normal. capability leadership in AI isn't static.

six months ago, claude 3.5 sonnet was arguably best. now gpt-5 has caught up and passed on some metrics.

six months from now? who knows.

the race isn't about any single moment. it's about sustained improvement and doing things right.

i can't share details (obviously), but we're not standing still. research continues. progress continues.

and importantly: we're focused on safety, not just capability. that's not just marketing—it's in every conversation, every meeting, every decision.

i want AI to go well for humanity. that's the real goal.

openai pushing forward means more pressure on everyone to improve. which could be good or bad depending on whether safety keeps pace.

i trust my company to prioritize the right things. i hope openai does too.

good news: you have excellent options. claude, gpt, and others are all strong.

pick based on what works for your use case. they're all getting better fast.

the AI race continues. we're in the middle of something historic. no pressure.