Urdu mein AI: Meta AI, Qalb-1.0 and Alif-1.0, and how to try each today
Three Urdu AI options you can use now: Meta AI in Urdu, the open Qalb-1.0-8B model and Alif-1.0-8B-Instruct. What each is, and how to start.
Urdu mein AI? Yes, and you don’t need to wait for it. There are already three options you can use today: Meta AI, which you can simply talk to in Urdu, and two open models you can run on your own computer.
1) Meta AI in Urdu: the easiest start
Meta launched Urdu support for Meta AI for users in Pakistan in October 2025 (Dawn, 27 Oct 2025). The Associated Press of Pakistan reported it as a launch by the Ministry of IT and Meta, and its headline calls the Urdu version “ALIF”. Don’t confuse that with Alif-1.0, the open model from Traversaal.ai in option 3 below.
How to start: just ask Meta AI in Urdu.
Good for: quick questions, drafting a short Urdu message, and checking how a sentence reads.
2) Qalb-1.0-8B: an open Urdu model you can run yourself
Qalb-1.0-8B is an open-weight Urdu large language model released in January 2026, built on Llama 3.1 8B. “Open-weight” means you can download the model and run it on your own machine, so your text doesn’t have to leave your computer.
Its authors report an overall score of 90.34 on Urdu benchmarks, against 87.1 for Alif-1.0. That is their own evaluation, published in their paper (arXiv 2601.08141), not an independent test. Treat it as a useful signal, not a verdict.
How to try it locally (if you already have Ollama installed):
ollama run enstazao/qalb:8b-instruct-fp16Check the model card on Hugging Face and the Ollama page (both linked below) for the available versions and hardware notes before you download.
3) Alif-1.0-8B-Instruct: open, Urdu-English, Apache 2.0
Alif-1.0-8B-Instruct is an open Urdu-English model by Traversaal.ai, released under the Apache 2.0 licence. It has GGUF builds, the format used by local tools such as Ollama and LM Studio.
Why the licence matters: Apache 2.0 is a permissive open-source licence. If you are a developer or a startup thinking about building on an Urdu model, read the licence on the model card first, so you know what you’re allowed to do with it.
Which one should you try first?
| If you… | Start with |
|---|---|
| Just want to ask questions in Urdu today | Meta AI |
| Want to keep text on your own computer | Qalb-1.0-8B or Alif-1.0-8B-Instruct |
| Are building a product and care about licensing | Read Alif’s Apache 2.0 licence and Qalb’s model card |
Kaam ki tip: three uses that save time at work
- Urdu customer replies. Paste the customer’s message and ask for a polite reply in Urdu. Then edit it in your own voice.
- Translating documents. Ask for an Urdu version of a short English notice, or the other way round.
- Summarising long Urdu news or WhatsApp threads. Paste the text and ask for five bullet points.
A simple prompt to copy:
Is customer message ka jawab Urdu mein likhein. Lehja polite aur professional ho, teen se chaar jumlay. Koi bhi qeemat ya tareekh apni taraf se na likhein.
(Write a reply to this customer message in Urdu. Polite and professional tone, three to four sentences. Don’t add any price or date on your own.)
The rule that matters most
Always double-check facts, names and numbers before sending. Any of these models can produce fluent Urdu that is still wrong. The AI drafts; you are responsible for what goes out.
FAQ
Is Qalb better than Alif?
On its authors’ own evaluation, Qalb scores higher (90.34 vs 87.1). We haven’t seen an independent comparison in our sources, so try both on your own work.
Do I need a powerful computer?
For Meta AI, no. For running Qalb or Alif locally, check the model card and Ollama page first, since local models need more memory than a normal app.
Sources
- Dawn, Meta AI launches Urdu version for users in Pakistan . Published 27 Oct 2025
- Associated Press of Pakistan (APP), Ministry of IT, META launch ALIF, an Urdu version for Meta AI
- Hugging Face, Qalb-1.0-8B-Instruct model card
- arXiv, Qalb paper, arXiv 2601.08141 . January 2026
- Ollama, enstazao/qalb
- Hugging Face, Alif-1.0-8B-Instruct (Traversaal.ai)
Drafted with AI assistance from the sources listed above; every figure and link is checked against those sources before publishing. Spotted an error? Tell us. Discuss it on LinkedIn.