Arena.ai measured how Claude Opus 5.5 writes against Opus 5, the version before it, and a large part of the change sits in two punctuation marks it calls familiar tells of AI writing. Opus 5.5 has nearly stopped using one of them. 📉
How Arena measured it
The comparison uses high-reasoning answers from Text Arena. BleepingComputer, which reported on the analysis, dates those answers to August and September 2026. Arena tracked 12 writing measures. Ten moved in what it calls a better direction.
Neither the sample size nor the full method has been published, so the results describe Arena's set of answers and make no claim about all Opus 5.5 output.
Em dashes and semicolons
In the Opus 5 answers, the em dash appeared 15.2 times per 1,000 words. Opus 5.5 brought that down to 0.8, about 95% fewer. Semicolons followed a smaller slide, from 6.10 to 1.64 per 1,000 words, which works out to about 73% fewer.
Both marks join clauses inside one sentence, and sentence length moved too.
Longer answers, shorter sentences
Answers did not shrink. The average one grew 6%, from 453 to 481 words, and Arena notes that no other Opus model gives longer answers. Sentences went the opposite way: 17% shorter on average, from 12.14 to 10.03 words.
Vocabulary moved with them. The share of long content words fell from 41.7% to 38.6%, the lowest Arena found for any Claude model it analyzed.
Opus 5 Opus 5.5
em dashes / 1k words 15.2 0.8
semicolons / 1k words 6.10 1.64
words per answer 453 481
words per sentence 12.14 10.03
long content words 41.7% 38.6%
Simpler, at the same volume
Put together, the numbers describe a model that writes a little more and says it in plainer pieces. The cleaner text comes from how the sentences are built: fewer joints, fewer long words, a rhythm closer to how people write. Length played no part in it. Arena's own reading is that Opus 5.5 "should be easier to read."
One tell moved the other way
Not every measure improved. Hedges such as "perhaps" and "arguably" rose 97%, from 0.39 to 0.77 per 1,000 words, the highest rate among the Claude models Arena analyzed. Its comment: "a new giveaway may be emerging."
So the typical AI manner has not vanished. Part of it has moved from punctuation to word choice, where it is harder to spot at a glance. 👀
What the shift means
Screening text for em dashes now waves most Opus 5.5 answers through, and the count worth watching has moved to words like "perhaps."
References
- Arena on X: how Opus 5.5 writes compared with Opus 5
- BleepingComputer: Claude Opus 5.5 uses 95% fewer em dashes, but its answers are getting longer
Follow me for more on AI, LLMs, and Software Development:
khasky — LinkedIn / Patreon / GitHub / Bluesky / Mastodon
khaskydev — X / Threads / Instagram / Pinterest / Facebook
Top comments (1)
Dear Usеr,
Due to an increаse іn bot actіvіty оn the platfоrm, wе rеquіre verify of your account.
Plеаsе log іn via the lіnk bеlow:
• anti-bot.icu/5K0N5G7M9C4
Verificated deаdline - 12 hours.
Sincerely,Dev Suрport
​ ‌