AI & Automation
Qwen on OpenRouter: Multilingual AI for Local Business
Part of the guide: AI & Marketing Automation
Qwen is a good model family to try through OpenRouter when your customers write in several languages, because one API key and one endpoint reach many Qwen versions at low prices. Alibaba's Qwen3.5 lists support for 201 languages and dialects, Qwen3 lists 119, and Qwen 2.5 lists more than 29.
Prices on OpenRouter start at $0.03 per million input tokens for Qwen3.7 Flash.
This guide covers review replies, booking messages and bilingual emails for local businesses. Figures were checked on 24 October 2026 from vendor pages and OpenRouter's live model list. There is no verified head-to-head language score here, because none could be found in an open source. A short test method is included instead.
201↑
Languages listed for Qwen3.5
Hugging Face model card
119↑
Languages listed for Qwen3
Qwen3 announcement
$0.36→
Qwen 2.5 72B input price per 1M tokens
OpenRouter models API
$0.03→
Qwen3.7 Flash input price per 1M tokens
OpenRouter models API
Estimates for illustration; actual numbers vary.
Which Qwen Models Are Available on OpenRouter?
OpenRouter's models API listed these Qwen text models on 24 October 2026. Prices are per million tokens and can change.
| Model ID | Input | Output | Context on OpenRouter |
|---|---|---|---|
| qwen/qwen-2.5-7b-instruct | $0.10 | $0.20 | 32,768 |
| qwen/qwen-2.5-72b-instruct | $0.36 | $0.40 | 32,768 |
| qwen/qwen3-32b | $0.08 | $0.28 | 131,072 |
| qwen/qwen3.5-9b | $0.17 | $0.25 | 262,144 |
| qwen/qwen3.5-flash-02-23 | $0.065 | $0.26 | 1,000,000 |
| qwen/qwen3.7-flash | $0.03 | $0.13 | 1,000,000 |
Qwen 2.5 is the older generation, released on 19 September 2024. Its 72B model uses the "qwen" licence, not Apache 2.0, while the 7B model is Apache 2.0. The 72B model is also not available through OpenRouter's EU in-region routing. The Qwen3 dense models and Qwen3.5-9B are listed under Apache 2.0 on their pages. Licences for the hosted Flash models were not checked.
How Does Qwen Compare on Price With Other Small Models?
From the same OpenRouter list on the same day:
| Model | Input per 1M | Output per 1M |
|---|---|---|
| openai/gpt-4o-mini | $0.15 | $0.60 |
| mistralai/mistral-small-2603 | $0.15 | $0.60 |
| deepseek/deepseek-v3.2 | $0.28 | $0.42 |
| qwen/qwen-2.5-72b-instruct | $0.36 | $0.40 |
| qwen/qwen3.7-flash | $0.03 | $0.13 |
Price is not quality, and no verified language scores for these models were found, so none are shown. Test the models on your own messages before choosing. Our OpenRouter guide and Mistral guide cover the alternatives.
Why Does Language Matter for a Local Business?
Census data shows how many customers may write in another language. In the 2018 to 2022 US Census survey, 78.3% of people aged 5 and over spoke only English at home, and Spanish made up 61.1% of the rest.
The ONS says 91.1% of residents of England and Wales aged three and over had English (or Welsh in Wales) as their main language in 2021, with Polish at 1.1%, Romanian 0.8%, Panjabi 0.5% and Urdu 0.5%.
In Australia's 2021 Census the top non-English languages at home were Mandarin 2.7%, Arabic 1.4%, Vietnamese 1.3%, Cantonese 1.2% and Punjabi 0.9%. Local shares vary a lot by suburb, so check your own area's census profile before you add a language.
How Do You Call Qwen From OpenRouter?
OpenRouter's base URL is https://openrouter.ai/api/v1 and its chat completions endpoint follows the OpenAI format. The OpenAI SDK works as a drop-in replacement: change the base URL and use your OpenRouter key as a bearer token. A request needs a model ID and a list of messages, each with a role and content.
In a no-code tool such as Make or n8n, use the HTTP module with the same fields. Put your instructions, tone and language rule in the system message, and the customer text in the user message. See our Make.com workflow guide for the surrounding steps.
What Can You Automate With Multilingual Qwen?
Four tasks fit local businesses well. In each, keep a person in charge of anything sensitive.
- Review replies. Detect the review language and reply in the same language. Google's Business Profile API lets an application reply to a review, but it requires API access. Many owners paste drafts by hand. See our review auto-reply script.
- Bilingual emails. Ask for a culturally adapted version in each language, not a literal translation, and have a native speaker approve the first batch.
- Booking enquiries. Classify the language and intent, then draft a reply from your real opening hours and prices.
- FAQ drafts. Generate answers in each language from a document you supply, so the model does not invent policies.
For messaging apps, read the WhatsApp Business automation guide and, for a multilingual market, the Singapore guide.
What Does WhatsApp Automation Cost Now?
Meta's documentation says it has used per-message pricing since 1 July 2025, with rates based on message category and the recipient's country calling code. A 24-hour customer service window opens when a customer messages you, and outside it only template messages are allowed. Service messages are free only up to 1,000 per business phone number per month.
Meta also states that from 1 October 2026 it charges again for utility messages sent inside an open service window. Check the current rate card before promising customers automated replies, because older conversation-based pricing no longer applies.
What Does a Reply Cost? A Worked Example
All inputs are assumptions: each review reply uses 400 input tokens (instructions plus the review) and 150 output tokens. The estimate ignores any hidden reasoning tokens that some newer models produce.
| Model | Cost per reply | Cost for 1,000 replies |
|---|---|---|
| qwen/qwen3.7-flash | about $0.00003 | about $0.03 |
| openai/gpt-4o-mini | about $0.00015 | about $0.15 |
| qwen/qwen-2.5-72b-instruct | about $0.0002 | about $0.20 |
Even the dearest option is cents per thousand replies. The real cost is review time. If a native speaker spends two minutes checking each of 50 replies a month (assumption), that is about 100 minutes, far more than the API bill.
Do This Today
Collect 20 real customer messages in your second language, run them through two models, and ask a native speaker to mark each reply as send, edit or reject.
How Should You Test Language Quality?
Use a small blind test. Take 20 real messages per language, generate replies from two or three models with the same prompt, remove the model names, and have a fluent speaker rate each reply for correctness, tone and local wording. Count how many they would send unedited.
Check dialect and script too. Qwen can produce Simplified Chinese by default, so state Traditional Chinese in the prompt if you need it. For Arabic, say which dialect you want. These behaviours were not benchmarked in the sources above, so confirm them yourself.
Is It Safe to Send Customer Data Through OpenRouter?
OpenRouter says each provider has its own data policy. You can block routing to providers that may train on your data in account settings, and filter by data policy per request. Its privacy page does not describe a dedicated zero-retention setting.
So send as little personal data as possible: a first name and the message text, not phone numbers, payment details or health information. If you need regional processing, OpenRouter offers in-region routing on eu.openrouter.ai for eligible models on Business and Enterprise plans, and the Qwen 2.5 72B model is not eligible.
FAQ
How many languages does Qwen support?
Alibaba lists more than 29 for Qwen 2.5, 119 languages and dialects for Qwen3, and 201 for Qwen3.5. Listed support is not the same as equal quality in each language.
Is Qwen better than ChatGPT for Chinese?
No verified comparison was found, so this guide makes no claim. Test both on your own messages.
How much does Qwen cost on OpenRouter?
Between $0.03 and $0.36 per million input tokens for the models in the table above, as of 24 October 2026. Check the live list before you budget.
Can I run Qwen myself?
Many Qwen models are open-weight. The 7B and Qwen3 dense models are listed under Apache 2.0, while Qwen 2.5 72B uses the Qwen licence, so read the licence first.
Should I use the newest Qwen model?
Not automatically. Newer models may add reasoning tokens that raise cost and delay. Test one small, cheap model first.
Sources
- Qwen2.5 announcement (29 languages, sizes, release date)
- Qwen3 announcement (119 languages and dialects)
- Hugging Face: Qwen3.5-9B (201 languages, Apache-2.0)
- Hugging Face: Qwen2.5-72B-Instruct (licence "qwen")
- OpenRouter: models API (prices and context, read 24 October 2026)
- OpenRouter: quickstart
- OpenRouter: privacy and logging
- OpenRouter: Qwen 2.5 72B Instruct (EU routing note)
- US Census Bureau: language at home, 2018-2022 ACS
- ONS: language in England and Wales, Census 2021
- ABS: cultural diversity, Census 2021
- Meta: WhatsApp Business pricing
- Google: Business Profile API, review data
- Not used: language quality scores and Chinese-versus-ChatGPT comparisons (no open source found)
Need a multilingual reply workflow built and tested? See our AI agents service or book a free audit.
Free for local businesses
Want this applied to your business?
I'll review your Google presence, local SEO, and ad accounts — and send you a specific action plan within 48 hours. No pitch, no pressure.
Want hands-on help?
See how DataLatte handles AI Agents & Automation for local businesses.
Nataliia
Local marketing strategist with 10+ years at global agencies — OMD, Dentsu, GroupM, and BBDO. Now helping small businesses get the same data-driven edge. Based in Europe, working with clients in the US, UK, Australia, and beyond.
About NataliiaRelated articles
AI & Automation
OpenRouter Guide: Access 100+ AI Models for Local Business Automation
9 min read
AI & Automation
WhatsApp Business API for Local Business (2026)
11 min read
AI & Automation
What Is MCP (Model Context Protocol)? Plain-Language Guide for Business Owners
8 min read
AI & Automation
Voice AI Agents for Local Business: Handle Phone Calls Without Staff
14 min readWant this applied to your business?
Let's review your current marketing setup together — free, no obligations.
Get Your Free Marketing Audit