Vendor claims, captured 2026-10-23 08:57 (UTC+8) with curl (raw copies in runs/2026-10-23/tools/qwen3-4b-deslop/vendor/)

Blog: https://100toolkit.com/blog/deslop-model/ (Erik Markus Kannike, October 2026)
  "A small open model, fine-tuned on a MacBook Air, that writes without em dashes, "not X, but Y" or lists of three and needs no system prompt, and you can run it locally with Ollama."
  "With Ollama it's ollama run hf.co/erikkannike/qwen3-4b-deslop-v4-GGUF:Q4_K_M, a 2.5 GB download."
  "On 40 prompts it never saw in training, the slop score dropped from 121 to 12.5 per 1000 words, and 68% of answers score in the range of human writing. Em dashes went from 20.5 per 1000 words to zero."
  "It still answers 30 of 30 factual questions correctly and fits the requested genre and length as often as the original."
  Limitations (author): "The voice is flat"; "it still makes things up ... invented names and places are common"; "It also runs long".

Model card: https://huggingface.co/erikkannike/qwen3-4b-deslop-v4
  Table (original -> this model): slop score 121 -> 12.5; em dashes per 1000 words 20.5 -> 0.0; "not X, but Y" per 1000 words 7.8 -> 0.8; lists of three 17.3 -> 1.1; 30 factual and arithmetic questions 97% -> 100%; damaged answers 0% -> 5%.
  "The slop score comes from a rule-based detector with 26 traits (lab/slopscore.py in the lab repo)." The lab repo is not linked from the card or the blog; a web search on 2026-10-23 did not find it.

Price: free open weights on Hugging Face, no account needed for download (checked 2026-10-23).
GGUF files used (sha256 matches the Hugging Face LFS ids):
  erikkannike/qwen3-4b-deslop-v4-GGUF qwen3-4b-deslop-v4-Q4_K_M.gguf d37cb32418014f5ffe74b55ac4572b524ce769a68a18036c3c3b0824666f4565
  unsloth/Qwen3-4B-Instruct-2507-GGUF Qwen3-4B-Instruct-2507-Q4_K_M.gguf 3605803b982cb64aead44f6c1b2ae36e3acdb41d8e46c8a94c6533bc4c67e597 (original model, same quantisation)