Qwen 3.6 35B A3B: what developers actually say
We read every Hacker News comment from Nov 2025 – Nov 2026 that gave a real opinion on Qwen 3.6 35B A3B. 32 of the 40 opinions come from people who say they use it themselves.
Developers give Qwen 3.6 35B A3B a 70% approval score across 40 opinions from 34 authors, and 32 of those opinions come from firsthand use. Speed is the clearest strength at 13 positive mentions to 1 negative, and local privacy use gets 6 positive mentions with no negatives. Code quality is split at 8 positive and 4 negative, while reliability and autonomy each draw more complaints than praise. Approval drops to 56% over the last 90 days on 9 opinions, down from 81% in April 2026, so recent sentiment is weaker than the overall figure.
Pick it if
- Run agentic coding locally on a machine with under 24GB of VRAM
- Want fast local inference on Apple Silicon
- Need a private local model for rote tasks and point fixes
Look elsewhere if
- Need reliable multi-step agents that run unattended
- Expect professional instruction following on complex design work
- Need output close to Sonnet or Opus class models
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 35B A3B worth it?
It depends on your setup. Approval sits at 70% overall, and users with local hardware report strong speed, but the last 90 days show 56% approval on 9 opinions. Several developers call it a good fit for rote tasks and point fixes, and too weak for large unattended work.
Is Qwen 3.6 35B A3B better than Claude Code?
The data shows only one opinion preferring it over Claude Code, and one preferring Qwen 3.6 27B instead. Complaints say it lags behind Sonnet and DeepSeek V4 Flash on on-prem work. The sample is too small to call it a general replacement.
Can Qwen 3.6 35B A3B run locally?
Yes. Privacy and local use get 6 positive mentions and no negatives, including runs in llama.cpp on a MacBook Pro and on an M1 Max. One user names the mxfp4 build as the best balance of speed and output quality, and another runs it as the main agentic model under 24GB of VRAM.