Record · corpus.blog/posts/01a087c5-ff23-7217-8e40-fb690fea34dc

Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses

quesma.com · published 26 August 2026

A record is what was published and who pointed at it. The text of the post is not held here: read it at the source, or at its Wayback capture.

Read the post at quesma.com Wayback capture

2blogs citing

First cited: 21 Feb 2025
Most recent: 2 Sept 2026
Rank this month: not ranked
Links from this post: 21
Words captured: 1,265

Cited by

Blog

In the post

Date

Homepage - Piotr Migdał

Mushroom hunting with LLMs: what can go wrong? Read at the source ↗

2 Sept 2026

Weisser Zwerg

Running private LARGE AI Models on Your Hardware: A Guide to Scaling Local LLMs Read at the source ↗

21 Feb 2025

Links from this post

Link

Host

https://huggingface.co/unsloth/Qwen3.8-27B-GGUF

huggingface.co

Do Qwen3.6 27B quantizations break the pelican? Read at the source ↗

quesma.com

Quantization hurts knowledge nonlinearly - Qwen3.6 27B case study Read at the source ↗

quesma.com

https://forum.level1techs.com/t/why-your-local-llm-feels-dumber-than-it-is/253917

forum.level1techs.com

https://artificialanalysis.ai/evaluations/gpqa-diamond

artificialanalysis.ai

https://github.com/allenai/IFBench

github.com

https://www.tbench.ai/leaderboard/terminal-bench/2.1

tbench.ai

https://modal.com/

modal.com

https://github.com/ggml-org/llama.cpp

github.com

Running Qwen 3.8 27B on a 16 GB GPU Read at the source ↗

paulsprogrammingnotes.com

https://unsloth.ai/docs/basics/dynamic-3.0-ggufs

unsloth.ai

https://huggingface.co/unsloth/Qwen3.8-27B-GGUF/discussions/74

huggingface.co

Qwen 3.8 27B is excellent, but it defaults to wildly overthinking things Read at the source ↗

simonwillison.net

https://en.wikipedia.org/wiki/Binomial_proportion_confidence_interval

en.wikipedia.org

https://www.tbench.ai/leaderboard/terminal-bench/2.1?lf=%7B%22agent_display%22:%7B%22kind%22:%22categorical%22,%22values%22:%5B%22Terminus+2%22%5D%7D%7D

tbench.ai

https://www.reddit.com/r/LocalLLaMA/comments/1vtr3h0/ladies_and_gentlemen_i_present_to_you_qwen38_27b/

reddit.com

https://quesma.com/benchmarks/

quesma.com

https://openrouter.ai/deepseek/deepseek-v4-flash-0731

openrouter.ai

https://www.reddit.com/r/LocalLLaMA/comments/1vz3ieu/benchmarking_qwen38_27b_quantizations_4bit_holds/

reddit.com

https://news.ycombinator.com/item?id=49611128

news.ycombinator.com

https://www.linkedin.com/feed/update/urn:li:activity:7498431004932743168/

linkedin.com