Skip to main content
Back to blog
•2 min read

GPT-4o Is Cool But I'm Biased Now

AIopenaithoughts

openai announced gpt-4o on may 13th. the "o" stands for omni. it's multimodal. it's fast. it's impressive.

and i have to be honest about my reaction: i watched the demo and thought "yeah but claude is still better."

is this bias? probably. let me try to be objective anyway.

what gpt-4o does

real-time voice: talk to it naturally, with interruptions and back-and-forth. the demos looked magical.

multimodal native: text, audio, images, video—all in one model. not stitched together, actually unified.

faster: responses come quicker. feels more conversational.

free tier: the good stuff is available to free users. democratization of AI continues.

the honest comparison

where gpt-4o might be better:

  • voice interaction (claude doesn't have this yet)
  • speed (genuinely snappy)
  • availability (everyone can use it)

where claude 3 might be better (in my experience):

  • longer context (200k vs 128k)
  • thoughtfulness in responses
  • uncertainty acknowledgment
  • less "slop" (fewer generic, overly-formal responses)

where they're similar:

  • raw intelligence (benchmarks are close)
  • coding ability
  • general helpfulness

my bias disclosure

i'm about to work at anthropic. i'm not exactly neutral here.

but here's the thing: my preference for claude predates my job offer. it's part of why i applied.

still, you should factor in that i have skin in the game now.

what this means for the field

competition is good. openai pushing hard means anthropic has to keep pushing. vice versa.

users benefit from this race. we get better models, faster, cheaper.

the risk is that race dynamics lead to cutting corners on safety. openai has been... aggressive about speed. i hope the industry doesn't sacrifice care for velocity.

my honest take

gpt-4o is impressive. if you're not using any AI tools yet, it's a great entry point.

but if you ask me which model i prefer for my own work? still claude.

(yes, i'm biased. no, that doesn't mean i'm wrong.)

tried gpt-4o's voice feature. said "honestly i prefer claude." it handled the rejection gracefully. good sportsmanship.