TII Launches Falcon Reasoning: Best 7B AI Model Globally, Also Outperforms Larger Models
TII ·
Falcon vs. SILMA
TII's homegrown open LLM vs. Google's Arabic-tuned Gemma 2 developed with GCC partners — two paths to GCC AI.
🇦🇪UAE
Falcon
TII
Open-source multilingual LLM family from TII, Abu Dhabi. First Arabic-organization model to top the Hugging Face Open LLM Leaderboard.
Sizes:7B, 40B, 180B, Falcon 2 (11B), Falcon 3 (1B–10B)
🇦🇪UAE
SILMA
Google / GCC Partners
Arabic-focused LLM by Google built on Gemma 2 with instruction tuning, developed with GCC partners.
Sizes:1B, 9B
Comparison
| Aspect | Falcon | SILMA |
|---|---|---|
| Creator | TII (UAE government lab) | Google + GCC academic partners |
| Base architecture | Custom Falcon architecture | Google Gemma 2 |
| Arabic support | Via Falcon 3 multilingual training | Dedicated Arabic instruction tuning |
| Max size | Falcon 180B (general), Falcon 3 up to 10B | 9B |
| Sovereign AI angle | Fully UAE-developed, sovereign model | Google model tuned for GCC context |
| Best for | Sovereignty-sensitive deployments, general tasks | Arabic tasks with Google ecosystem integration |
Bottom Line
Falcon represents true GCC AI sovereignty — built from scratch by a UAE institution. SILMA offers Google's optimization quality with Arabic focus. Both are open-source.