Observed arrival · 2026-10-27
Caphlon: a 27B model in 6.87 GiB
A downloadable GGUF build of Qwen 3.8 27B, presented as a way to run a large language model within 8 GB of memory.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- People running local language models on 8 GB machines
- Worth noticing
- The page labels HumanEval as 96.7% in one place and 68.4% in its comparison table.
Field notes
The page describes a GGUF build made through vocabulary merge-table reduction and hybrid-aware layer pruning, rather than distilling a smaller student model. It supplies a llama.cpp command and says the build also works with Ollama; listed backends include Metal, Vulkan, CUDA, and CPU vector extensions. The project reports different throughput figures for an M3 Max and a base M1 MacBook Air, while its benchmark table compares the build with 27B and 70B models.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
◎NicheUnusually specific use
One card from the complete issue