Tawkeed Arabic Benchmark Leaderboard
Comprehensive Arabic LLM evaluation across 22 skills
None defined yet.
Empowering Arabic-first AI — on your terms, on your premises.
Tawkeed is a Saudi-native enterprise Edge AI platform purpose-built for Arabic. Aligned with Saudi Vision 2030, Tawkeed delivers state-of-the-art AI capabilities that run entirely on-premises — no internet required, no data leaves your walls.
Our Arabic-first large language models cover 18 Arabic dialects and are optimized for the unique linguistic, cultural, and regulatory needs of the region.
| Product | Description |
|---|---|
| Tawkeed Chat | Enterprise conversational AI assistant with deep Arabic fluency and domain expertise |
| Tawkeed Meeting | AI-powered meeting transcription, summarization, and action-item extraction for Arabic |
| Tawkeed Slides | Intelligent Arabic presentation generation from documents and briefs |
| Tawkeed Code | Arabic-aware coding assistant for developers — code generation, review, and documentation |
| Tawkeed Agents | Autonomous AI agents that execute multi-step workflows across enterprise systems |
| Custom Solutions | Tailored AI deployments fine-tuned to your organization's data and domain |
Purpose-built hardware for private, on-premises AI inference:
Arabic-first large language models designed for enterprise use:
| Metric | Value |
|---|---|
| Cost reduction vs. cloud AI | 70% |
| Response latency | <10ms |
| Uptime SLA | 99.99% |
| Arabic dialect coverage | 18 dialects |
| Metric | Value |
|---|---|
| Organizations served | 500+ |
| Active users | 50,000+ |
Government · Financial Services · Healthcare · Enterprise · Retail · Manufacturing · Education
| Model | Parameters | Description |
|---|---|---|
| tawkeed-20b | 20B (MoE) | Arabic text generation, conversational AI, summarization |
| tawkeed-ocr | 2B | Arabic OCR and document understanding (vision-language) |
| tawkeed-embedding | 568M | Multilingual embeddings for Arabic semantic search and RAG |
We maintain the Tawkeed Arabic Benchmark — 970 questions across 22 skill categories evaluating Arabic language models: