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

Article

TL;DR

Alibaba claims 27B dense model matches flagship coding benchmarks — Opus-level performance, runnable locally.

Key Takeaways

• 27B dense model claims to match frontier coding performance on published benchmarks
• If benchmarks hold, this dramatically lowers the cost of flagship-level code generation
• Self-reported benchmark methodology draws immediate skepticism; independent evals pending

Discussion

Top comments:

• [cgeier]: Beating Opus 4.5 with 27B is extraordinary — their own benchmarks, hard to believe
• [pama]: Community immediately asking for real-world hands-on impressions, not benchmark tables

Discuss on HN