OpenAI has to really cook at this Dev Day
OpenAI has to really cook at this Dev Day.
- GPT-6 Sol has been a fiasco
- The current subscription doesn’t even last 2 days with Astra
- And they want to cut its API-equivalent usage in half
I get that now that they’re “pacing the frontier,” they’re going to focus a lot more on model efficiency. We’ve already seen that with Anthropic.
The idea would be for the Codex sub to work more or less the same even with half the API credits, because the new models will be quite a bit cheaper and more efficient by default.
In short: we’re cutting the tokens you get in half, but in return the models also use half as many, and we’re giving you extra features you didn’t have before.
I think that’s the simplified idea.
And on paper, it makes sense.
This will probably help their margins a lot, let them introduce that new $500 tier they want to add, and put them in a better position for a potential IPO.
My only concern is GPT-6 Sol.
If Sol had good performance, none of this would worry me too much. But to make it this much cheaper, they shipped a 💩.
I get that’s why at this Dev Day they’ll also have to present extra stuff that gives users more value without necessarily requiring more tokens.
If they do something similar to what Anthropic did with Opus 5.5 and Sonnet 5.5, I think that’s a fair deal.
If they repeat the Sol bullshit, it’s a rip-off.
For now I’m skeptical, but I trust them.
We’ll see what they announce at Dev Day.
