How to Use Qwen2.5-VL 72B Instruct in Muse Code
Use Qwen2.5-VL 72B Instruct to extract data from scanned documents and explain charts. It can also locate objects with bounding boxes or points, making it useful when an answer needs a position in the image.
Prompts to try with Qwen2.5-VL 72B Instruct in Muse Code
Copy an example into Muse Code, or ask in your own words.
Use Qwen2.5-VL 72B Instruct to inspect the screenshots in this project's visual tests. Label the reference Image A and the current render Image B, list visible layout differences, then find the components responsible.
Use Qwen2.5-VL 72B Instruct to extract the invoice images in the test fixtures into JSON records. Follow the project's existing invoice schema and use null for unreadable values.
Use Qwen2.5-VL 72B Instruct to read the chart screenshot in the issue. List its title, axis labels, legend, and visible values, then compare them with the chart configuration in the project.
Use Qwen2.5-VL 72B Instruct to locate the Save button in the failing test screenshot. Return a labeled bounding box, then inspect the related test and component before proposing a fix.
Use Qwen2.5-VL 72B Instruct to inspect the object-counting fixture. Locate every visible bottle first, report the total, and compare the result with the expected annotation.
How to set up Qwen2.5-VL 72B Instruct in Muse Code
Pick the way that suits you. Each takes about two minutes.
Use Spiral's MCP server
- Install Muse Code
curl -fsSL https://dev.meta.ai/install.sh | bash
- Ask Muse to add the server
Start muse and ask:
Add the Spiral MCP server at https://api.spiral.ai/mcp
- Sign in
In your terminal, run:
muse mcp login spiral
Use the Spiral skill
- Send Muse Code this message
Your agent reads Spiral's skill and sets itself up.
set up https://spiral.ai/SKILL.md
- Sign in or paste a key
Approve Spiral in the browser, or give the agent a key from the console.
Use the Spiral CLI
- Install and sign in
Installs the Spiral CLI, signs you in and adds its skill to your coding agents.
npx -y spiral-cli@latest init
Call the Spiral API
- Create an API key
In the console, under API keys.
- Run Qwen2.5-VL 72B Instruct
One request; the result comes back by polling or a webhook.
curl https://api.spiral.ai/v1/runs \ -H "Authorization: Bearer $SPIRAL_API_KEY" \ -H "Content-Type: application/json" \ -H "Idempotency-Key: $(uuidgen)" \ -d '{ "provider": "openrouter", "model": "qwen/qwen2.5-vl-72b-instruct", "input": { "messages": [ { "role": "system" } ] } }'
Then paste one of the prompts above: Muse Code finds Qwen2.5-VL 72B Instruct on Spiral, shows the price and runs it.
Qwen2.5-VL 72B Instruct providers and prices
Spiral runs Qwen2.5-VL 72B Instruct through OpenRouter. Open a card for today's price, with every fee included.
Up to $0.005362 per run at the default inputs. $1.04 per 1M input tokens, $1.30 per 1M output tokens.
Settings you can ask for
Ask Muse Code for any of these in your prompt.
| Setting | Allowed values | Default | Description |
|---|---|---|---|
| Seed | a whole number | ||
| Top k | a whole number from 0 | ||
| Top p | 0 to 1 | ||
| Logprobs | on or off | ||
| Max tokens | 1 to 115200 | The most output tokens to generate. | |
| Temperature | 0 to 2 | ||
| Top logprobs | 0 to 20 | ||
| Presence penalty | -2 to 2 | ||
| Frequency penalty | -2 to 2 | ||
| Repetition penalty | 0 to 2 |
Every API field of Qwen: Qwen2.5 VL 72B Instruct is on its tool page: open its card above.
What can you do with Qwen2.5-VL 72B Instruct in Muse Code?
Turn screenshots and document images into evidence for code changes with Qwen2.5-VL. Ask your coding agent to connect visible labels, layouts, or extracted records to the relevant project files.
Invoice extraction
Turn invoice images into records containing the supplier, invoice number, date, line items, and total. Mark unreadable values for review.
Form transcription
Extract printed labels and completed entries from form images into a consistent table.
Chart explanations
Read chart titles, axes, legends, and visible values, then explain the comparison in plain English.
Screenshot review
Inspect interface screenshots for visible labels, layout differences, and the locations of named controls.
Multilingual text collection
Transcribe signs or document images in languages covered by Qwen's OCR training, preserving the original script.
Object annotation
Locate specified objects in images and return labeled bounding boxes or points for review.
Visual counting
Identify each visible instance of a requested object before returning a count and locations that a reviewer can inspect.
More on Qwen2.5-VL 72B Instruct and Muse Code
Overview
Qwen2.5-VL 72B Instruct in other agents
Questions about Qwen2.5-VL 72B Instruct in Muse Code
Can Muse Code use Qwen2.5-VL 72B Instruct?
Yes. Add Spiral to Muse Code as an MCP server, and Muse Code can find and run Qwen2.5-VL 72B Instruct like any built-in tool.
How much does Qwen2.5-VL 72B Instruct cost in Muse Code?
Up to $0.005362 per run at the default inputs. $1.04 per 1M input tokens, $1.30 per 1M output tokens. You pay from your Spiral balance as you go, with no subscription, and a failed run is not charged. Your Muse Code plan is separate.
Do I need an OpenRouter account to use Qwen2.5-VL 72B Instruct in Muse Code?
No. Spiral runs Qwen2.5-VL 72B Instruct on its own OpenRouter account. You sign in to Spiral once and never paste a key.
How long does it take to set up Qwen2.5-VL 72B Instruct in Muse Code?
About two minutes: install Muse Code, then ask Muse to add the server, then sign in.
Does Muse Code wait while Qwen2.5-VL 72B Instruct runs?
Qwen2.5-VL 72B Instruct runs in the background. Muse Code starts the run, checks on it and gives you the text, json when it's ready.
What is Qwen2.5-VL useful for?
Use it to extract information from document images, interpret charts, and locate objects. Good starting tasks include turning an invoice scan into structured data or explaining a chart's labels and values.
Can Qwen2.5-VL extract tables and forms?
Yes. It supports structured extraction from scanned tables, forms, and invoices. Specify the columns you want and ask it to mark unreadable entries instead of guessing.
Which languages does Qwen2.5-VL's OCR training cover?
Qwen names French, German, Italian, Spanish, Portuguese, Arabic, Russian, Japanese, Korean, and Vietnamese in its multilingual OCR training data. Treat that list as training coverage, not a promise of equal accuracy across languages.
What context limit does Qwen publish?
The model card describes a default configuration of 32,768 tokens. It also warns that YaRN context extension affects spatial and temporal localization, so a longer configuration should not be assumed to preserve location accuracy.
How should I prompt Qwen2.5-VL to count objects?
Ask it to locate each object first, then report the total. Qwen uses this detect-then-count approach in its reported evaluation; check the returned locations against the image.
Start using Qwen2.5-VL 72B Instruct in Muse Code
One Spiral account for every tool your agent needs. No subscription; you pay per run.
Paste this into Muse Code:
set up https://spiral.ai/SKILL.md


