Opus 5.5 vs. Codex
- Right now the most powerful coding model is Anthropic's [Opus 5.5](/en/news/anthropic-claude-opus-5) — it reads and writes code way faster and better than OpenAI's models, and the subscription lasts longer too.
- That said, computer use in Codex is still miles ahead of Claude's — it handles almost everything on the web: sign-ups, admin panel setup, orders. I've stopped suffering over bad web interfaces — I just don't see them anymore. And it's nicer to talk to as well.
The compute race
A few weeks ago it looked like Anthropic had serious compute problems while OpenAI had practically unlimited capacity. Today, with Opus 5.5, Claude has enough for everyone — meanwhile OpenAI still won't let new users buy the $200 plan (20x the usage of the base tier).
The big question: will OpenAI unveil a new coding model tomorrow, at its annual Devdays conference, that beats Opus 5.5 — or not.
The AI assistant market
There are already more than a hundred players in the market. A lot of people I know in the US use Instinct — everything runs through iMessage and texts. And ads for Meta's (Facebook's) Muse are plastered all over the New York subway.
The whole scene is also buzzing about how the movement to rein in AI overlaps, membership-wise, with the friends of Aella, the sex-worker-and-data-scientist — but I recently found out my oldest daughter and her friends read this channel, so I'll skip the recap.
Did I miss anything important?