I’m all-in on OpenAI and Codex. Some people asked my why after my X post, and I do not have a one-line answer.
I believe the latest Anthropic and OpenAI models are both strong. For most people, benchmark gaps are almost invisible in real use (who can tell the difference between 79.7% and 80.3% from the trenches?).
The bigger benefit comes from everything around the model: skills, project structure, and tool access. If that layer is solid, most top models can produce good work. You could even use Gemini. But please, don’t.
My problem with Opus is not that it is weak. It is that it is brilliant, then charges into poor decisions. I spend more time cleaning up side effects than producing value, and that trade kills momentum.
Codex 5.3 became very, very good. It is not always the fastest, but when I run 5 sessions in parallel, I am the bottleneck anyway. What matters is judgment: solving the user’s problem without breaking something else in the codebase. That is where it wins for me.
I think OpenAI still pays for its mainstream image. They made choices I see as distractions, and Sora is the clearest one for me: interesting demo, expensive detour, GPU bonfire, good to prepare an IPO for engagement, not what I need in my life. But that does not change model quality.
There is also a social layer people pretend is purely technical. Opus became a badge in parts of Tech Twitter, especially in alpha-male, terminal-first circles that like to signal distance from the mainstream (and don’t get me started on the condescension when they have to talk to “normies”).
Mid-February 2026, that badge is mostly inertia. For code, Codex is already in the same league, including for people who live in the terminal.
At some point, this becomes Android vs iOS all over again. You pick an ecosystem and build depth, or you keep switching and stay in permanent setup mode. If you keep hopping, you lose compounding gains because your workflow never stabilizes.
Last part is personal.
I was spending close to $1,000 per month at Anthropic, and I am not paying that to get life lessons about what I should or should not do with it. I already have my mother for that, and she costs less. I just heard, as I write theses lines, that they have updated their ToS to make sure it was now really forbidden to use anything else but their tools with a subscription, and that's OK.
That corporate policy chapter could be very long but I'm sure you already get the idea.
They can run their company the way they want and I can spend my money the way I want. OpenAI now gets my vote.