GPT-5: First Impressions After 24 Hours

**Update** OpenAI have acknowledged many of these shortcomings and have re-enabled GPT-4o. Additionally Sam Altman revealed there was a bug with the GPT-5 rollout causing its internal model selector to consistently pick the least powerful model. Pretty rocky start for GPT-5 all things considered, but I've seen some reports that the research mode can do incredible work.

OpenAI announced GPT-5 yesterday. The pricing is reasonable. The results look promising. Giving state of the art AI to the masses (with limited free plan usage) will allow those resistant to AI to see what really good AI is like without paying for it.

That being said, this was not the revolutionary launch we were pitched.

Along with this launch, every other model has been forcefully removed from the web interface.

Some compared it to Apple’s launch of the iPhone 16 and I agree. Watching a 70+ minute keynote for an incremental update that could’ve been an email did not feel like a good use of time.

For the past few OAI launches I've been afraid they would introduce tech capable of replacing most knowledge worker jobs with zero configuration leading to huge economic instability. Thankfully we aren't there yet.

My Suspicion of What’s Really Going On

• 4o and o3 are expensive loss leaders designed to drive adoption.
• GPT-5 is a cost-saving algorithm aimed at making providing ChatGPT cheaper to end users
• OpenAI is doubling down on practical use cases. AI is a tool, not your friend.

Coding Experience

Anecdotally, I prefer o3. These greenfield "make a program from scratch" examples are cool, but they aren’t representative of real world use.

Unleashing GPT-5 and expecting it to fix a broad codebase is still not productive.

If anything, this is an incremental improvement. Their stats - when presented with a coherent y-axis support this.

Missing 4o’s Personality

4o felt like it had a real sense of humor. Heavy users of 4o are extremely upset, and cancelling their subscriptions left and right. If you relied on ChatGPT for nuanced communications purposes, creative writing, or as a thoughtful conversation partner, prepare for disappointment.

Also it still loves those pesky emdashes… They say it produces better writing–it does not.

Rate Limiting

"Plus" subscribers are now officially (even more) second class citizens. Users report quickly running out of requests on GPT-5.

Altman claimed he was delivering “a team of PHDs in your pocket” that would elevate any person to the level of the most effective person in history. Maybe one day. But not today.