Personally how I see it, benchmarks aside, Fable really is a tier above both Opus and Codex. Still makes mistakes, still needs supervision, but much less, and with a good feedback loop is able to achieve a lot independently. If OpenAI had something tucked away, they'd have probably already released it, right? I'm worried as this big of a difference means that we're probably going to speed towards a big price increase, at least until some competitors catch up. But if Anthropic really just built a moat and they don't slow down, the market is going to be much different, and the price of frontier model(s) will follow. Even if you want to go for open source, having enough power to run 1T param models means a non trivial investment for many companies out there.
If Anthropic "won" they'd not 2x the price as well as potentially lock it out of monthly plans. Opus is already on the pricier side. Sonnet hasn't been updated in a long while. Haiku is even worse.
Right now, the only thing this gains is Anthropic has a better model but at 2-5x the cost. Is it worth it? That's not "winning". That's a tradeoff.
Is that a better way to think of Anthropic's pricing structure?
I'm not saying there's no value. Just like there's expensive tea and cheap tea. The expensive tea doesn't "win" though, which in this context is referring to defeating all the competitors like a >50% marketshare scenario.
And it's not like everyone is sitting still. Every competitor is evolving rapidly. Whether it is ability or pricing.
That's why I'm posting this, my expectation (based on previous behavior) is that there should be a response from OpenAI. Considering it's missing I'm worried there's some secret sauce that others are missing.
You don't need 1000s of architects to build a building.
Just like companies don't have everyone as staff engineers.
IF as I explained this was the mid-level engineer (on price) beating every existing staff engineer by miles, then that's winning.
This is not.
> That's why I'm posting this, my expectation (based on previous behavior)
In history, it is still within the time range that OpenAI has reacted and released a new model, so no, nothing new.
Yeah, I'm rooting for this otherwise we're in for a rough ride in the next couple of months.