How many em dashes each AI lab put in its model launch posts over the last two months: 22 posts from 8 labs. Each mark below is one em dash. Bracketed ranks are ties.
Average em dashes per launch post. Anthropic averages as many as the rest of the field.
Method. Each post's article body was downloaded and its U+2014 characters counted by script; navigation, footers and cookie banners are excluded. En dashes (–) and hyphens don't count. Dashes inside customer quotes and table cells do. On the posts both cover, this method matches the original tally exactly: Fable 5 & Mythos 5 = 26, Opus 5.5 = 18, Opus 4.7 = 17, Sonnet 4.5 = 14, Fable 5.1 = 12.
Window. Model launch posts on each lab's own site from 10 Sep to 10 Nov 2026. Just outside: GPT-6 Astra (3 Sep, 5 dashes) and Claude Fable 5.1 & Mythos 5.1 (1 Sep, 12). Labs without a launch post on their own site (Qwen, MiniMax, Z.ai) aren't included.
Per 1,000 words. Posts under 500 words are marked short; one dash moves their rate a long way. xAI's pages have no article tag, so their word counts include some page chrome.
| Launch post | Lab | Date | Em dashes | Words | Per 1k |
|---|