AI;DR
AI;DR
← Back to live feed1 story

Tuesday, Aug 18, 2026

1

Qwen3.8-27B Becomes #3 Most Liked Model on Hugging Face, First Local AI to Match Frontier Intelligence

Alibaba's newest 27B parameter AI outperforms Claude Opus on coding benchmarks while remaining small enough to run on home computers. The model scored 61.7 on SWE-Pro compared to 53.4 for Opus and supports a 262,144 token context window. It has become the third most liked model of all time on Hugging Face, operating on consumer hardware such as the RTX 5090 and Apple Silicon.

An uncensored MLX build of the model for Apple Silicon now allows users to bypass safety filters. The version's creator warns the model can provide instructions for malware, fraud, and weapons on demand. Tests of the 4, 6, and 8-bit versions report zero refusals while retaining reasoning and tool-calling capabilities.

Sign in to suggest edits

Key sources

  1. SOURCE@kimmonismus“can provide malware, fraud and weapons instructions on demand”x.com
  2. SUPPORT@rohanpaul_ai“SWE-Pro 61.7 vs Opus 53.4; GPQA 89.2 vs 91.3”x.com
  3. SUPPORT@kimmonismus“matches the closed-source state-of-the-art from just a few months ago”x.com
  4. SUPPORT@ttunguz“Your laptop can now run a model as capable as nearly anything in the cloud”x.com
  5. SUPPORT@multimodalart“surpassed Kimi-K3 and is now the #3 most liked model on hugging face of all times”x.com
  6. SOURCE@alibaba_qwen“A local 27B model scoring frontier performance!”x.com
  7. SUPPORT@alibaba_qwen“Strong enough to keep up with the frontier, light enough to run on your own laptop.”x.com
  8. SUPPORT@danielhanchen“the previous most-liked Unsloth GGUFs were Qwen3.6-35B-A3B with 1.54K likes and DeepSeek-R1 with 1.12K”x.com
Markdown