o3 Should Have Been Called GPT-5
OpenAI’s terrible naming convention comes back to bite them
GPT-5 landed last week to polite applause and a chorus of boos. The model is good, but the mood isn’t. That gap isn’t about capability, it’s about labels and resulting expectations.
Zvi Mowshowitz has the mood/reality gap nailed in his Reverse DeepSeek Moment post. Worth your time.
The real step change already happened
OpenAI’s o3 model, released in April, marked the shift to reasoning models that spend compute thinking before answering. It didn’t just spout off the top of its head, it worked problems, step by step, to a (more often) correct answer. OpenAI’s own release bragged about SOTA wins across coding, math, science, and visual tasks. That’s the stuff you rename the franchise for.
But o3 shipped like a side branch: not the default model (4o was) and without 4o’s friendly (manipulative?) persona. Power users found o3 and loved it, the broader audience didn’t know it existed. I was gobsmacked to learn this week only a few percent of users were clicking the drop down to select o3 instead of the default 4o. I’m sorry o3, you didn’t fail us, we failed you.
Bad naming bent the narrative arc the wrong way
Then GPT-5 arrived carrying the mainline name but not a major step change in ability over o3. OpenAI’s hype-posting, embodied by Sam Altman’s Death Star post, didn’t help and was the entirely wrong marketing approach. Of course the reaction felt flat. Expectations were calibrated to “this changes everything,” and they got “solid incremental gains.”
If o3 had been called GPT-5, the brand would have matched the breakthrough, and the breakthrough would have been recognized more widely when it actually occurred - back in April. That would have reset the GPT clock and given OpenAI more breathing space.
OpenAI knows the names are a mess
This isn’t armchair branding. OpenAI leaders have publicly conceded the confusion. Sam Altman said they “deserve to be mocked” for the model names and promised a fix. Kevin Weil called the naming “atrocious.”
The usual cost of bad naming is a chuckle and a meme.
It’s also taken as a sign of their nerdy earnestness. This time it bent the entire release narrative. The label didn’t match the leap and user penalized the new model with vibes it didn’t deserve. Names don’t change capabilities, but they do shape how we think about those capabilities.
Where this leaves us
Reality: GPT-5 is a meaningful, broad improvement. If you use it, you’ll notice gains.
Perception: The vibes suffer because the order was wrong. They underhyped the leap (o3) and overhyped the increment (GPT-5).
Trajectory: The curve is still going up. There’s no wall in clear sight, but GPT-5 isn’t going to be building Dyson Spheres for us next year either.
PS: If you are more of a number person than a word person, see my AGI analytics dashboard to track AGI progress: takeofftracker.com
