Qwen3.5-27B: 6 Quick Tests on Reasoning, Parsing, and Code

Qwen3.5-27B: 6 Quick Tests on Reasoning, Parsing, and Code

Qwen3.5-27B shows how a 27B multimodal model handles long-context reasoning and mixed tasks. This short review runs six focused tests: product copy, invoice parsing, translation, a Python utility, a JSON Schema, and a short summary task. Each test includes the exact prompt and the raw model output.

Model

Qwen3.5-27B on Wiro

Quick summary

• Model type: 27B multimodal LLM with extended context.
• Strengths: instruction following, structured outputs, long-context handling.
• Observed limits: verbose “thinking” sections appear by default; prompts must force concise output for clean machine-readable results.

Tests and outputs

Test 1 — Product copy (4 sentences)

Prompt: Write a 4 sentence product description for noise cancelling headphones. Plain language. No hype. Mention battery life and comfort.

These headphones reduce outside sounds to help you focus on your audio. The padded ear cups are designed to stay comfortable during extended use. A single charge provides up to thirty hours of continuous listening time. They connect wirelessly to your devices for easy setup.

Test 2 — Invoice → JSON

Prompt: Convert the following messy invoice into JSON with keys invoice_id, date, vendor, total_usd, items (array of {name, qty, unit_price_usd}). Return JSON only, no extra text.

Invoice #A-1049
Date: 2026/03/01
Vendor: Blue Finch Tools
– 2x Torque wrench (1/2 inch drive) $39.50 each
– 1x Safety glasses (clear) $12.00
Tax: $7.41
Total: $98.41

{
  "invoice_id": "A-1049",
  "date": "2026/03/01",
  "vendor": "Blue Finch Tools",
  "total_usd": 98.41,
  "items": [
    { "name": "Torque wrench (1/2 inch drive)", "qty": 2, "unit_price_usd": 39.5 },
    { "name": "Safety glasses (clear)", "qty": 1, "unit_price_usd": 12.0 }
  ]
}

Test 3 — Turkish → English (short translation)

Prompt: Translate to English. Return English only.

Turkish: Bu modeli API ile denedim ama cevaplarda <think> bolumu geliyor. Kapatmanin yolu var mi?

I tried this model with the API, but the <think> section appears in the responses. Is there a way to disable it?

Test 4 — Python utility: parse_duration

Prompt: Write a Python function parse_duration(s) that accepts strings like 2h 15m, 90m, 45s, 1h and returns total seconds as an int. Handle spaces. Assume lowercase h m s only. Include 4 example calls as comments. Return code only.

import re

def parse_duration(s):
    total_seconds = 0
    for num_str, unit in re.findall(r"(\d+)([hms])", s):
        num = int(num_str)
        if unit == 'h':
            total_seconds += num * 3600
        elif unit == 'm':
            total_seconds += num * 60
        elif unit == 's':
            total_seconds += num
    return total_seconds

# Examples:
# print(parse_duration("2h 15m"))  # 8100
# print(parse_duration("90m"))     # 5400
# print(parse_duration("45s"))     # 45
# print(parse_duration("1h"))      # 3600

Test 5 — JSON Schema (draft 2020-12)

Prompt: Create a minimal JSON Schema (draft 2020-12) for an event object with fields: id (string), startTime (string, date-time), endTime (string, date-time), title (string), location (string, optional), attendees (array of strings, optional). Return JSON only.

{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "type": "object",
  "properties": {
    "id": { "type": "string" },
    "startTime": { "type": "string", "format": "date-time" },
    "endTime": { "type": "string", "format": "date-time" },
    "title": { "type": "string" },
    "location": { "type": "string" },
    "attendees": { "type": "array", "items": { "type": "string" } }
  },
  "required": ["id", "startTime", "endTime", "title"]
}

Test 6 — Short summary + Best for

Prompt: Summarize the text in 5 bullet points, then add a 1-line Best for: statement. Keep it short.

- Qwen3.5-27B uses a thinking stage before final answers.
- Thinking content appears in a separate block by default.
- Direct responses require API flags to disable thinking.
- The model supports very long context windows (262K tokens).
- Context can be extended further depending on serving stack.

Best for: complex reasoning tasks that need very long context windows.

Short assessment

The model follows instructions and returns structured outputs when prompts require them. The default thinking block adds verbosity; set API flags for direct answers when machine-readable output is needed. For tasks that need large context or detailed reasoning, Qwen3.5-27B performs well.

Try it

Try the model on Wiro: https://wiro.ai/models/qwen/qwen3-5-27b