Theo. Opus 5.5 takes half of all T3 Code prompts
The first model in Theo's coding product to cross a 50% share of traffic, by his own account.
Theo says Opus 5.5 has crossed a line no other model has in his coding product. In a post on X he reports that it is the first model to break the 50% traffic threshold in T3 Code, and that literally half of all prompts now go to it.
Opus 5.5 is the first model we've seen break the 50% traffic threshold in T3 Code. Literally half of all prompts go to Opus
That is one number from one product, posted by the person who runs it, with a screenshot attached. It is not a benchmark and it is not an independent measurement. What makes it interesting is the kind of signal it is. Benchmarks measure a model against a fixed task list that labs can and do optimize for. A traffic share measures what people pick when they have other options sitting in the same menu and nothing stopping them from switching mid task.
What the number does not say
Theo did not give a time window, a prompt count, or the breakdown for the models sitting underneath Opus 5.5. He also did not say whether the split reflects defaults, pricing inside the product, or users actively choosing. A model that is first in the dropdown gets traffic for reasons that have nothing to do with whether it is good. Until someone posts the rest of the chart, the honest reading is narrow. Half the prompts in one product go to one model, and no model in that product has done that before.
Why it lines up with the rest of the week
Opus 5.5 has been the model people keep running experiments on, including Theo's own earlier work with it. A usage share is a different genre of evidence from those one off runs. An engineer finishing a hard port in a single session tells you the ceiling. A traffic split tells you what the median user reaches for on an ordinary Tuesday, which is usually the more boring and more useful fact.
For working engineers, the thing to watch is whether other tool builders publish comparable splits. One vendor's internal mix is an anecdote at scale. Two or three independent products showing the same concentration would be something closer to a finding. Nobody has posted that yet.