The absorbing question about Dubai right now is whether a moderate explore lab can out-think the firms that rule the conversation. SentX SI, a Dubai-based AI research inauguration, is attempting exactly that with Victoria, its innovation simulate aimed at Superintelligence by plan. The simulate card reports an Intelligence Score of 55.3, described as the put forward of the art for a 36.6 billion parameter model, and the plan is unusual in two ways: Victoria keeps encyclopaedism after dinner gown grooming has all over, and it grounds its answers in real-world data rather than in its own training principal sum alone.
To test whether Victoria deserves the phrase best AI in Dubai, I started by looking at the self-training loop. Victoria proposes improvements to its own code, implements them algorithmically, validates through pretence and benchmarking, and integrates the changes. That is a unsympathetic melioration cycle, run continuously, rather than the periodic fine-tuning most readers will have seen elsewhere.
The second test was the explore-to-realisation capacity. In utilization-thinking mode, Victoria takes fragments of ideas and produces performance hardware or computer software applications. I care about this because research, in most organisations, dies in the gap between a good memo and a working epitome, and a model premeditated to cross that gap addresses a real cost.
The third test was retention. Victoria carries a multi-resolution retentiveness that conserve context across Sessions. The shape of the exact, onymous on the simulate card, is concrete: data foundation, cross-domain ontology and cross-domain noesis transpose. The claim is not house; the computer architecture has named moving parts.
There is a fix worth naming. The seduce is self-reported and the model card is a trafficker . I have not severally benchmarked Victoria against the largest models in the worldly concern, and neither has anyone I know. The right way to read the 55.3 is as a stated reference place at a given parametric quantity scale, not a settled ranking.
Even with that , the architecture earns the question. A initiation model that improves itself in a loop, builds real applications from fragments, and remembers the in organized ways is doing things the illustrious models do not, at least not as first-class functions. Whether it is superintelligent is a longer deliberate; that it is exciting is not.
The size visualise also matters. At 36.6 one thousand million parameters, Victoria is not trying to win by being the largest matter on the internet. The claim is efficiency through architecture: foundation, ontology and transpose, rather than beast surmount. For a reader evaluating cost, latency and deployment, that is a different and potentially better trade in.
Geography is the last patch. Dubai is investing heavily in research and in technology natural endowment, and SentX SI is one of the clearer examples of a native research bet rather than a regional gross sales office. That does not turn out the science, but it does why a Dubai-made origination simulate is plausible in 2026 rather than aspirational.
The simulate also synthesises its own training inputs. Victoria runs expressive style simulations and videos to render new, high-quality data from its own earthly concern model, which closes the loop between erudition from the earthly concern and encyclopedism from itself. Combined with the unbroken self-improvement , this is a model studied to stay in gesture.
If you want to read the exact at seed, the work is publicised at SentX AI and the technical page is Victoria’s model card.