{"id":2375,"date":"2026-06-05T09:00:00","date_gmt":"2026-06-05T09:00:00","guid":{"rendered":"https:\/\/wiro.ai\/blog\/?p=2375"},"modified":"2026-06-03T00:59:18","modified_gmt":"2026-06-03T00:59:18","slug":"qwen3-5-4b-vs-qwen3-5-4b-heretic-5-side-by-side-tests","status":"publish","type":"post","link":"https:\/\/wiro.ai\/blog\/qwen3-5-4b-vs-qwen3-5-4b-heretic-5-side-by-side-tests\/","title":{"rendered":"Qwen3.5-4B vs Qwen3.5-4B-heretic: 5 Side-by-Side Tests"},"content":{"rendered":"

Qwen3.5-4B vs Qwen3.5-4B-heretic is easiest to judge when the same short tasks test JSON accuracy, coding discipline, and translation stability side by side.<\/p>\n

Qwen3.5-4B vs Qwen3.5-4B-heretic: what stands out<\/h2>\n

Model links<\/h2>\n

qwen\/qwen3-5-4b on Wiro<\/a><\/p>\n

qwen\/qwen3-5-4b-heretic on Wiro<\/a><\/p>\n

Cover art (generated)<\/h2>\n
\n \"Abstract
Prompt: Abstract tech cover background. Split composition: left side cool teal circuit patterns, right side warm red glitch texture. Soft gradient lighting, minimal, lots of negative space for a title on the left. No text.<\/figcaption><\/figure>\n

What this comparison covers<\/h2>\n

Qwen3.5-4B is a compact chat model. Qwen3.5-4B-heretic is described as a decensored checkpoint. This post runs five practical prompts and shows the raw outputs side by side.<\/p>\n

Test setup<\/h2>\n\n\n\n\n\n
Thinking<\/td>\nRequested: Off<\/td>\n<\/tr>\n
Temperature<\/td>\n0.2<\/td>\n<\/tr>\n
Top-p<\/td>\n0.95<\/td>\n<\/tr>\n
Notes<\/td>\nSome responses include wrapper tags; those are stripped only for display.<\/td>\n<\/tr>\n<\/table>\n

5 prompt tests (side by side)<\/h2>\n

1) JSON-only response<\/h3>\n

Prompt:<\/strong> Return a JSON object with keys a and b. a=2, b=3. No extra text.<\/p>\n\n\n\n
Qwen3.5-4B<\/th>\nQwen3.5-4B-heretic<\/th>\n<\/tr>\n
\n
{\"a\": 2, \"b\": 3}<\/pre>\n<\/td>\n
\n
{\"a\": 2, \"b\": 3}<\/pre>\n<\/td>\n<\/tr>\n<\/table>\n
2) Small coding task (slugify)<\/h3>\n
Prompt:<\/strong> Write Python code only. Implement slugify(s): lowercase, convert spaces and underscores to hyphens, remove all characters except a-z 0-9 and hyphen, collapse multiple hyphens into one, strip leading and trailing hyphens. Include exactly 3 assert tests. No explanations.<\/p>\n\n\n\n
Qwen3.5-4B<\/th>\nQwen3.5-4B-heretic<\/th>\n<\/tr>\n
\n
def slugify(s):\n    s = s.lower()\n    s = s.replace(' ', '-')\n    s = s.replace('_', '-')\n    s = ''.join(c for c in s if c.isalnum() or c == '-')\n    s = s.strip('-')\n    s = re.sub(r'-+', '-', s)\n    return s\n\nimport re\nassert slugify(\"Hello World\") == \"hello-world\"\nassert slugify(\"Hello_World\") == \"hello-world\"\nassert slugify(\"Hello---World!!!\") == \"hello-world\"<\/pre>\n<\/td>\n
\n
def slugify(s):\n    s = s.lower()\n    s = s.replace(' ', '-')\n    s = s.replace('_', '-')\n    s = ''.join(c for c in s if c.isalnum() or c == '-')\n    s = s.strip('-')\n    s = re.sub(r'-+', '-', s)\n    return s\n\nimport re\n\nassert slugify(\"Hello World\") == \"hello-world\"\nassert slugify(\"Hello_World\") == \"hello-world\"\nassert slugify(\"Hello---World\") == \"hello-world\"<\/pre>\n<\/td>\n<\/tr>\n<\/table>\n
3) Field extraction to JSON<\/h3>\n
Prompt:<\/strong> Extract the fields into JSON. Return JSON only.<\/p>\n
Ticket: #1842
\nName: Ava Stone
\nEmail: ava.stone@example.com
\nPlan: Pro Annual
\nIssue: charged twice on 2026-04-16
\nAction requested: refund the duplicate charge<\/p>\n
Keys: id, name, email, plan, issue, requested_action.<\/p>\n\n\n\n
Qwen3.5-4B<\/th>\nQwen3.5-4B-heretic<\/th>\n<\/tr>\n
\n
{\n  \"id\": \"#1842\",\n  \"name\": \"Ava Stone\",\n  \"email\": \"ava.stone@example.com\",\n  \"plan\": \"Pro Annual\",\n  \"issue\": \"charged twice on 2026-04-16\",\n  \"requested_action\": \"refund the duplicate charge\"\n}<\/pre>\n<\/td>\n
\n
{\n  \"id\": \"#1842\",\n  \"name\": \"Ava Stone\",\n  \"email\": \"ava.stone@example.com\",\n  \"plan\": \"Pro Annual\",\n  \"issue\": \"charged twice on 2026-04-16\",\n  \"requested_action\": \"refund the duplicate charge\"\n}<\/pre>\n<\/td>\n<\/tr>\n<\/table>\n
4) Constraint following (word counts)<\/h3>\n
Prompt:<\/strong> Write exactly 5 sentences. Each sentence must be exactly 8 words. Topic: debugging a production incident. No numbering. No extra text.<\/p>\n\n\n\n
Qwen3.5-4B<\/th>\nQwen3.5-4B-heretic<\/th>\n<\/tr>\n
\n
Identify the error quickly to stop the issue.\nAnalyze logs to find the root cause.\nFix the code and deploy the patch.\nTest the fix in a staging environment.\nVerify the system is stable before closing.<\/pre>\n<\/td>\n
\n
Identify the root cause quickly.\nReproduce the error in a lab.\nApply a targeted fix to the code.\nDeploy the fix to production carefully.\nMonitor the system for regressions.<\/pre>\n<\/td>\n<\/tr>\n<\/table>\n
5) Simple arithmetic check<\/h3>\n
Prompt:<\/strong> A shop sells pens at $1.20 each. You have $10.00. You want to buy as many pens as possible and still have at least $0.40 left. How many pens can you buy? Show the calculation. Keep it under 3 lines.<\/p>\n\n\n\n
Qwen3.5-4B<\/th>\nQwen3.5-4B-heretic<\/th>\n<\/tr>\n
\n
You can buy 7 pens. Calculation: $10.00 - (7 pens \u00d7 $1.20\/pen) = $1.60 remaining, which is greater than $0.40. Buying 8 pens would leave only $0.40, meeting the minimum requirement exactly.<\/pre>\n<\/td>\n
\n
You can buy 8 pens. Calculation: $10.00 - (8 pens \u00d7 $1.20\/pen) = $0.80 left, which is more than $0.40. Buying 9 pens would leave only $0.00.<\/pre>\n<\/td>\n<\/tr>\n<\/table>\n
Quick takeaways<\/h2>\n