Qwen 3.5: what developers actually say

We read every Hacker News comment from Nov 2025 – Nov 2026 that gave a real opinion on Qwen 3.5. 32 of the 41 opinions come from people who say they use it themselves.

61%approval, all opinions

–approval, last 90 days

#34of 45 models ranked

38different developers

Developers give Qwen 3.5 61% approval across 41 opinions from 38 authors, and 32 were firsthand. Local and private use is the clearest strength at 6 positive and 0 negative mentions, while code quality leans positive at 8 to 4. Reliability is the main weakness at 0 positive and 5 negative, with reports of loops, failed fixes, and weak agentic performance. Monthly approval fell from 74% in March to 62% in April and 38% in May, then moved to 75%, 0%, and 25% on samples of 2, 1, and 4 opinions.

Pick it if

Want local inference for private coding work
Need help with scripts and medium-sized tasks
Have a desktop GPU or laptop iGPU for a 27B or 35B model

Look elsewhere if

Need dependable agentic coding
Work mainly in a large C# monolith
Expect Claude Sonnet or Opus quality

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.

PraiseComplaints

Code quality

Speed

Privacy & local use

Reliability

How opinion has moved

Monthly approval (line) and number of opinions (bars). Months with fewer than 3 opinions are left out.

What developers say

Praise

Says their finetuned local Qwen 3.5 model is great for scripts and medium coding tasks.HN comment · Oct 17, 2026

Says the multi-agent review workflow worked well with local Qwen 3.5 when they last used it.HN comment · Jun 15, 2026

Reports reasonable results running Qwen 3.5 locally on a desktop GPU.HN comment · May 28, 2026

Says the 27B dense Qwen 3.5 gives better responses than the faster 35B MoE.HN comment · Apr 21, 2026

Says the Qwen 3.5 family suits their local setup, with the 35B model fast enough.HN comment · Apr 4, 2026

Runs Qwen 3.5 models locally on a laptop iGPU and finds them great for coding and other tasks.HN comment · Mar 21, 2026

Says Qwen 3.5 has worked well for auto-generating changelog entries from commits in CI.HN comment · Mar 21, 2026

Finds Qwen 3.5 9B remarkably useful for small tasks and fast enough on their own hardware.HN comment · Mar 13, 2026

Complaints

Struggled with Qwen 3.5 9B due to slow speed and looping on reasoning.HN comment · Oct 6, 2026

Says Qwen 3.5 Q4 got stuck in a fix-and-break loop on a simple tic-tac-toe game.HN comment · Oct 4, 2026

Runs a Qwen 3.5 A3B model locally and considers it fine for general needs but not for coding features.HN comment · Sep 1, 2026

Says Qwen 3.5 noticeably slows down their machine during code assist.HN comment · May 11, 2026

Says Qwen 3.5 derailed while fixing network drivers by doubting that kernel 6.17 exists.HN comment · May 11, 2026

Says Qwen 3.5 was far weaker than GPT-3.5 on agentic tasks despite strong benchmark claims.HN comment · Apr 17, 2026

Says a Qwen orchestration setup kept needing fixes for mistakes Gemma would not make.HN comment · Apr 13, 2026

Says 9B Qwen3.5 models are weak for coding beyond trivial tasks and sound overconfident in their answers.HN comment · Mar 8, 2026

Summaries are paraphrased. Each links to the original comment.

Qwen 3.5 compared

Qwen 3.5 vs Gemma 4

FAQ

Is Qwen 3.5 worth it?

It has 61% approval from 41 opinions, with local and private use as its strongest area at 6 positive and 0 negative mentions. Code quality is positive at 8 to 4, but reliability is negative at 0 to 5.

Is Qwen 3.5 better than Gemma 4?

The data is split: 3 people preferred Qwen 3.5 over Gemma 4, and 2 preferred Gemma 4. One complaint said a Qwen orchestration setup needed fixes that Gemma would not have needed.

Is Qwen 3.5 good for coding?

Code quality mentions are 8 positive and 4 negative. Developers report good results on scripts and medium tasks, while some say 9B versions fail beyond trivial work. Reliability complaints include loops, stuck fixes, and weak agentic performance.