Oman Is Building Its Own AI Language Model. Why?

Oman's government is training its own AI language model, called Maeen, on local data. That sounds like a ministry project a long way from your front desk. But it answers the same problem your phone line runs into every day: software that handles textbook Arabic and stumbles on the way your callers actually talk, at the cost of an operator's time or a lost booking.

What Maeen actually is

The Ministry of Transport, Communications and Information Technology (MTCIT) announced Maeen at COMEX on 8 September 2025 as a model meant to reflect Omani language and culture. A language model is the engine inside chat assistants: software that has read huge amounts of text and learned to understand and write language.

According to the national AI programme's head, the pilot was built on more than 3,000 local datasets with 60 government institutions taking part. The first phase was expected to benefit around 20,000 government employees. At a May 2026 session, the ministry showed it drafting official letters, translating large files and summarising long reports. MTCIT also says it keeps data secure inside Oman.

What it is not: a public app, or something you can plug into your switchboard. When this article was written, no release date for the public or for businesses had been published. Continuing to develop it is a priority in the national AI programme's second phase, which runs from 2026 to 2030.

Why build one when global AI already speaks Arabic

Big global AI models learn Arabic mostly from what people write online. Most of that is Modern Standard Arabic, the formal Arabic of news and official documents, plus the dialects with the biggest online presence. Omani Arabic, which sounds different in Salalah, Sur, Nizwa and along the Batinah coast, is a thin slice of that pile.

So a generic system writes a formal letter well. Then a caller from Ibri asks about a refund in a hurry, and it guesses. The second reason is control. Training on government data inside the country keeps that data under Omani law and on Omani servers.

When the government decides generic Arabic AI is not good enough for its own staff, it is telling every business that the way people here speak is worth getting right.

The same gap, on your phone line

Take Saif, a composite example: he runs operations for a three-branch polyclinic group in Seeb. A caller rings at 7pm: "Baghait appointment bacher 'ind doctor al asnan, fi waqt ba'd al maghrib?" She wants a dentist tomorrow, any time after maghrib.

His old IVR (interactive voice response, the press-one-for menu) made her listen to six options first. The voice bot he trialled after that was trained on general Arabic. It caught "appointment", missed the rest and asked her to repeat herself twice. She either waited for an operator or hung up. What finally changed Saif's numbers comes further down. First, the sums.

What misunderstanding costs a busy desk

None of that shows up as a line in the budget. It hides inside operator overtime, flat booking numbers and callers who quietly try the clinic down the road. For more on why dialect trips machines up, see why most AI still misunderstands Omani Arabic.

What the state is signalling to businesses

• Dialect becomes the baseline. Once the state builds for local speech, "it understands Arabic" stops being enough of an answer from any vendor.
• Where data lives matters. Maeen is pitched as keeping data inside Oman. Expect customers and regulators to ask the same about your call recordings and transcripts.
• The focus is moving to business. On 22 October, the government launched the second phase of the National Digital Economy Programme, covering 2026 to 2030. It shifts attention from government services to businesses, targets RO 300 million in technology investment and encourages AI in marketing, stock and demand forecasting.
• There are ground rules. MTCIT's policy on safe and ethical use of AI sets principles such as fairness, transparency and accountability. It was issued as a circular rather than a law with fines, but it shows what the state expects.

How Saif got his line ready for local speech

Saif did not wait for Maeen, which is a government tool and not a product for his switchboard. Instead, he tested vendors against his own callers:

1. He pulled 50 real recorded calls from a busy week (callers were told calls are recorded), mixing Omani Arabic, other dialects and English.
2. He played them to each vendor's system and counted how many it understood the first time, with no "please repeat".
3. He replaced the press-one menu with a conversation that handles booking, timings and directions. Anything sensitive goes to an operator with a written summary, so nobody repeats their story.
4. He asked each vendor in writing where recordings and transcripts are stored.
5. Every week, he reviews the calls the system got wrong and feeds them back.

Transfers to the operator desk dropped, and his team spent their shifts on the calls that genuinely needed a person. This is the job CustomerCare.OM's Omani Arabic voice agents are built for, but the same 50-call test works on any vendor.

What this means for you

You do not need to follow the national AI programme closely. What you do need to know is that dialect understanding is becoming the standard people expect. If your IVR or voice bot still makes callers rephrase things in formal Arabic, you are paying for it in operator time and lost bookings every day.

Before you renew a contract or sign a new one, test with your own recordings, not the vendor's demo. Ask where the data is stored. And give your callers a fast route to a human, along with a summary, so the person who picks up already knows what the call is about.

The bottom line

Oman is building its own language model because generic Arabic AI does not hear the country the way it speaks. Your phone line faces the same test every day. Measure how often it fails, fix it with tools tested on your own callers, and the money you save is the reason it matters.