Opus 5.5 takes half of all T3 Code prompts
Two teams report the same swing toward Anthropic in the same week.
Theo says Opus 5.5 is the first model to break the 50% traffic threshold in T3 Code. Literally half of all prompts in his coding tool now go to Opus. That is a single tool's internal number, not a market share report, but it is the sharpest version of a shift several builders have been describing, and we covered the first read on it earlier.
The second data point lands the same week and comes from somebody else entirely. dax says Claude is his team's most used model for the first time since the beginning of the year, and credits Opus 5.5 directly. Two independent teams reporting the same swing is about as close to a signal as you get before anyone publishes real numbers.
for the first time since the beginning of the year claude is our team's most used model
Theo's scorecard
Theo put rough rankings on the models he uses. Opus 5.5 gets 9/10 on code capability, 8.5/10 on prose and readability, 8/10 on price and 9/10 on understanding intent, which he notes are improvements across the board. GPT-6 Astra lands at 7.5/10 on code capability and 8/10 on prose, but 3.5/10 on price and 3/10 on understanding intent. His summary of Astra is that it is better at code but worse where it matters.
He is explicit that this is a reversal of his own position. In July 2026, by his account, Anthropic had the best code models but the gap was small, they were slow and expensive, and a $200 subscription could be burned through in a day. Preferring OpenAI models made sense because they were faster, cheaper and more pleasant to use.
By September 2026 he says the gap is much bigger, Opus 5.5 is fast and surprisingly cheap, and the $200 subscription feels nearly limitless. He now says you can kill a $200 OpenAI plan in a few hours and that the $500 plan goes even faster with UltraFast.
The business read
These are vibe rankings from one person who ships a coding tool, plus one team lead's usage observation. Neither is a benchmark. What makes them worth reading is that both are people paying the bills, and both moved in the same direction in the same quarter. Theo adds the caveat himself, saying OpenAI can come back from here, but that it is looking rough right now.
Preferring OpenAI models for code makes almost no sense.
Steal the numbers, not the vibes
One email a day. Under three minutes. Real MRR, real pricing, real launches from founders building with AI.

