Qwen

A C#-compiler-flavored fragment (“can be converted to foreach”) that became the most famous “polysemantic” glitch token: the model perceives it as a different word every time. It exists in cl100k, Llama-3 and Qwen vocabularies because the tokenizers share BPE merges.

A Dutch postal-code API fragment that is glitchy across three separate model families — a good demonstration that glitch tokens transfer whenever tokenizers share BPE merges.

A Roslyn/C# API identifier preceded by a tab character — a code-corpus artifact that is under-trained in Llama 3.

Membership-promo boilerplate from a business-data platform (Tianyan Shengyitong). Listed by the Zhihu community as a Qwen dirty token.

Reprint/copyright boilerplate from CSDN-style blogs, with a trailing space. Listed by the Zhihu community as a Qwen dirty token.

The second member of the useRal triplet (cl100k id 89472), a C#-flavored identifier fragment; the same string also exists in the Qwen2 vocabulary.

A Dutch postal-code API fragment. The signature under-trained token of the cl100k lineage, “migrating” into Llama-3 and Qwen vocabularies via shared BPE merges.

A .NET Selenium documentation fragment beginning with a literal tab — one of the smallest-embedding tokens in the Llama-3 vocabulary.

The Dutch word for “receiving clients at home”.

A high-frequency fragment of Chinese web “for more information” disclaimers — an under-trained long token in the Qwen3.5 vocabulary (id 128904).