Qwen/Qwen3.6-27B · Hugging Face

Article

TL;DR

Qwen3.6-27B drops on HuggingFace with same-day unsloth 4-bit GGUF quant for local runs.

Key Takeaways

• 4-bit GGUF quantization from unsloth available immediately — runs on consumer hardware
• Claims to beat Claude Opus 4.5 on coding benchmarks using non-standard test configurations
• If verified independently, local 27B models at flagship quality reshapes the build-vs-API calculus

Discussion

Top comments:

• [throwaw12]: Unsloth 4-bit quant available same day — community moves fast on local deployment
• [GodelNumbering]: Terminal-Bench 2.0 scores use non-standard harness configs — benchmark validity questionable

  Terminal bench 2.0 rules explicitly disallow modifying timeouts or resources available.

• [cgeier]: Beating Opus 4.5 with a 27B model — hard to believe, pending independent validation

Discuss on HN