evidoria

← Back to browse

Good practice Imported

Karnak — Egypt's Sovereign Arabic Large Language Model Powering Government AI Applications

Egypt · Cairo · See the Egypt profile · See the Cairo profile

Evidence: Descriptive / self-reported Top 62% 53/100 · Ask Evidence Copilot about this practice

Egypt's Ministry of Communications launched Karnak, a ~40-billion-parameter Arabic-English large language model built on Qwen3, as sovereign public infrastructure powering government tools for education, legal assistance, translation and call-centre auditing.

~40 billion parameters
Model size
Karnak — Egypt's Sovereign Arabic Large Language Model Powering Government AI Applications

Details

Maturity
Pilot
Promoter
Ministry of Communications and Information Technology (MCIT) / Egypt's National Council for Artificial Intelligence
Period
Feb 2026–ongoing
Keywords
digital sovereignty, generative AI infrastructure, public administration

Context

On 11 February 2026, Egypt's Ministry of Communications and Information Technology launched Karnak, the country's first large-scale, state-built Arabic-English language model, positioned as sovereign digital infrastructure.

Activities

Karnak-40B (~40 billion parameters, built on Alibaba's Qwen3-30B-A3B-Instruct architecture) powers flagship applications: SIA (AI tutor), AcQua (call-centre interaction auditing), Torgoman (translation), BelMasry (colloquial-Arabic NLP) and an AI legal assistant, and can run inside local data centres for data sovereignty.

Results

The government describes Karnak as ranking among the top-performing Arabic-language models in its size class.

Conclusions

As of mid-2026, public information is limited to capability benchmarks and the list of intended applications; no independent data on production accuracy, uptake, or measured efficiency gains has been published.

Implementation
Indicative cost
Very high (> €5M) — Not disclosed; training and hosting a ~40B-parameter sovereign LLM is a substantial infrastructure investment.
Time to results
Long (> 3 years) — Launched 11 February 2026; applications unveiled alongside it.
Staffing & skills
Ministry of Communications and Information Technology, Egypt's National Council for Artificial Intelligence

Conditions for success

  • Built on an established open architecture (Qwen3) rather than training from scratch
  • Local data-centre hosting framed as a digital-sovereignty safeguard

Common failure modes

  • No independent production accuracy or uptake data published

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful