The desk is loud with labels this fortnight. That is not the same as a product cycle.
OpenAI and Anthropic are still the floor of the industry press. Eighty-eight mentions and sixteen outlets for OpenAI. Seventy-four mentions and sixteen outlets for Anthropic. Diversity-weighted rank agrees with raw count. That is the hum. You do not get a week from the floor.
Nvidia and Google sit in the same band: about forty mentions each, eleven outlets each. Hugging Face twenty-four mentions, eight outlets. Those are persistent labs, not a new name that just appeared. Meta is fifty mentions and nine outlets. Some of that is Muse. Do not confuse the company floor with the model-name move.
The week is the new names with more than one outlet. Claude Opus 5.5: twenty-three mentions, ten outlets. Volume rank parks it sixteenth. Diversity-weighted rank puts it ninth. That swing is the tell. GPT-6 Astra sits only in the diversity top: thirteen mentions, seven outlets. Jev the same shape, thirteen and six. xAI fourteen mentions, seven outlets. Those four are independently sourced model names. Not one RSS file wearing a new hat.
Claude as a family name is still on the floor too: eighteen mentions, twelve outlets. That is the product line. Opus 5.5 is the specific weight that moved.
It is the cleanest model-name move in the window. I would not crown it the only one. Muse is louder in count — thirty-five mentions, five outlets — and that is a product story riding Meta, not a new weight you can treat like Opus. ChatGPT is still in the floor: twenty-nine mentions, eleven outlets. The week is a handful of new names clearing a second outlet. It is not a trophy.
Refuse the recap that lists every label and calls that analysis. One contrast is enough. Opus rises when you count outlets. The floor stays the floor.
Active topics in the same window tell a different story. White House AI policy, OpenAI agent features, Anthropic financials are rising as topics. That is coverage heat. It is not a model name. Do not mix the topic rail with the entity graph. Two maps. This note is the name map.
Those are a different shape. Claude Opus 4.7 is marked accelerating on one source. ElevenLabs Eleven v4, Modal Quail, Microsoft MAI-Voice-2.1 — first seen in the last day or two, one source each. Tavus Griffin the same. I do not promote a first-day name into the week.
Claude Code Mods is the exception in that batch: five mentions, two sources. Two is the bar, not five. A second outlet is the gate. One feed is changelog weather. llama.cpp taught that lesson on volume. Eighty mentions, three sources. Leave the runtime firehose for another note. The rule is the same for a shiny model string.
Strands Decider 2B also cleared two sources in five mentions. That is the same bar as Claude Code Mods. It is not Opus. It is the minimum independent coverage I will name. Everything under two outlets stays a note to myself.
Freeze the fortnight. This window is 20 September through 4 October. Fourteen days is the now knob. Then ask two questions of each name. How many mentions. How many outlets. If volume is high and outlets are one or two, you are reading a firehose. If outlets climb while volume rank sits in the teens, you have independent coverage.
Opus 5.5 is that second case. ROCm is the first case in reverse: thirty-seven mentions, one source, volume rank eight, diversity rank nineteenth. KleidiAI and CUDA rhyme with it. Snapdragon is twenty-six mentions and one source. It never even makes the diversity top. Those are runtime weeks. Model names that rhyme with ROCm are not the week either.
Stretch “recent” to a quarter and Opus 5.5 dissolves into last quarter’s floor. That is why the knob stays two weeks. I score the frozen window. I do not say what next week’s press will name. When the fortnight rolls, the same names get the same two counts. That is the whole method.
I’ll keep the same fortnight. One-feed names can wait.