GLM-5.3 averages 7.4 em dashes per response; Claude models use almost none
ArtificialAnlys · x · 2026-11-05
Artificial Analysis measured average em dashes per response across models in AA-PersonalityBench:
- GLM-5.3 leads at 7.4 per response, followed by GLM-5.3-Flash (6.9), Muse Wonder (6.9), and Kimi K3 (6.7)
- Claude Fable 5.1 averages 6.4, but Claude Opus 5.5 uses fewer than one every 40 responses
- Near-zero users include GPT-3.5 Turbo (0.00), Claude Haiku 5.5 (0.00), Claude Sonnet 5.5 (0.00–0.01), and Llama 4 Maverick (0.01)
The stat works as a quantified stylistic fingerprint: model families differ sharply in punctuation habits, and newer Claude models appear to have dropped the em dash entirely.
Related event: GLM-5.3 Tops Em Dash Usage Count While Claude Barely Uses Any(2 posts)→
More from Fun
- Lenny Rachitsky: Tesla FSD subscription is the one I'll never cancel — lennysan · 2026-11-05
- Rider finds a Toy Story alien buddy charging in their Waymo — reed · 2026-11-05
- Instagram becomes YouTube (90-min videos, Apple TV) while YouTube becomes Instagram (algo feed) — omooretweets · 2026-11-05
- Coinspect researcher escapes AWS Firecracker sandbox with AI agents' help — evilsocket · 2026-11-05
- Gary Marcus: my 3-year-old LLM predictions on moats and price wars all came true — GaryMarcus · 2026-11-05
- Screenshot appears to show Gemini's internal reasoning steps rendered to users — can_a_bus · 2026-11-05