Qwen 3.8 27B just embarrassed models many times its size.
A 27B model you can run locally beat Claude Opus 4.6 Max on 15/19 launch benchmarks. And the benchmarks aren't even the most interesting part. The numbers: → DeepSWE 1.1 jumped from 13.3 → 42.2 → That's a 217%
Doing some live testing not finished but current state .
Do you think its useable ? 2 b70s Qwen 3.8 27b FP16 Deepseek harness
Qwen 3.8 27B Q4 is now running on an RTX 4060 with just 8GB VRAM and a 64,000 token context window using Unsloth's new IQ4_XS quant at 14.6GB on disk.
→ Prefill at 150 tokens per second, decode at 5 tokens per second via native MTP → Only 25 GPU layers offloaded to stay within
GLM 5.3 vs Gemini 3.7 Flash vs Qwen 3.8 vs Grok 4.6
You have reached the end of the archive
All of qwen38