Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model

· ai llm open-source · Source ↗

Article

TL;DR

Qwen’s 27B dense model claims Opus-class coding performance running on consumer hardware.

Key Takeaways

  • Q4_K_M quant runs at ~16.8GB on M5 Pro; fits 32GB RAM machines
  • Local frontier-quality coding narrows the moat of paid API providers significantly
  • Benchmark comparisons to Opus are disputed; wait 2 weeks for community verification

Discussion

Top comments:

  • [simonw]: Runs well on M5 Pro 128GB; pelican SVG better than Opus 4.7

    The pelican is excellent for a 16.8GB quantized local model… I like it better than the pelican I got from Opus 4.7 the other day.

  • [jedisct1]: Found 8/10 security bugs in benchmark with zero false positives
  • [originalvichy]: Wait two weeks; early models often have hidden backend bugs

Discuss on HN