Qwen 3.6 27B: what developers actually say
We read every Hacker News comment from Nov 2025 – Nov 2026 that gave a real opinion on Qwen 3.6 27B. 63 of the 76 opinions come from people who say they use it themselves.
Developers give Qwen 3.6 27B 76% approval across 76 opinions from 58 authors, 63 of them firsthand. Praise centers on code quality (21 positive, 4 negative) and local privacy (15 positive, 0 negative), with 6 positive cost comments and no cost complaints. The trend is down: 90-day approval is 56%, and monthly approval fell from 93% in June to 50% in September and 0% in October on just 2 opinions. Complaints cluster around reliability (5 negative, 4 positive), slow speed, and weak performance in long agent runs.
Pick it if
- Want a capable coding model you can self-host
- Need private inference on your own hardware at low cost
- Do basic editing, scripting, and debugging on a local GPU
Look elsewhere if
- Need reliable long agentic coding sessions
- Need fast throughput for interactive coding agents
- Need frontier-level coherence on complex or novel work
Strengths and weaknesses
Each row counts opinions that mentioned the topic. Bars to the right of the line are praise, bars to the left are complaints.
How opinion has moved
What developers say
Praise
Complaints
Summaries are paraphrased. Each links to the original comment.
FAQ
Is Qwen 3.6 27B worth it?
It depends on the workload. Overall approval is 76%, and users report strong code quality and local privacy. Recent approval is 56% over 90 days, so test it on your own tasks before committing.
Is Qwen 3.6 27B better than Claude Sonnet?
The data is split. One user preferred it to Sonnet 4.5, and one report says self-hosting it matches Sonnet 4.6 at low cost. Others found the local model far behind Sonnet and Cursor's default, so results vary by setup and task.
Can Qwen 3.6 27B run locally?
Yes. Local and privacy comments are 15 positive and 0 negative. Reported speeds range from 19 to 21 tokens per second down to about 10 tokens per second on an M5 MacBook Pro, and some users call that too slow for coding agents.