Dubai-founded artificial intelligence company AHOY Labs ranked eighth among 26 systems in EnterpriseRAG-Bench, a public test of how accurately enterprise AI platforms retrieve, assess and answer questions from organisational data. The company said its result placed it ahead of products from OpenAI, Amazon, Microsoft, Google and NVIDIA.

EnterpriseRAG-Bench assesses AI search and retrieval systems using a simulated corporate archive of 500,000 records. The test is designed to reflect challenges faced by organisations working with internal data, including emails, chat messages, files, tickets and meeting transcripts.

Its questions can require systems to connect information held in several documents, deal with conflicting records or recognise when the archive does not contain an answer. The benchmark evaluates both the accuracy of responses and the precision of the documents returned to support them.

Published results gave AHOY Labs an overall score of 74.95, with 79.6 per cent of questions answered correctly. OpenAI File Search recorded a score of 61.03 and a correct-answer rate of 69.8 per cent, according to the results. Microsoft Azure AI Search scored 48.42, while Google Vertex AI Search scored 41.87.

Precision results

AHOY Labs reported its largest advantage in precision, which measures the number of irrelevant documents returned by a system when it produces an answer. Its platform returned an average of 1.03 irrelevant documents per response, compared with 15.7 for OpenAI File Search.

“We built AHOY to prove that sovereignty and world-class performance are not a trade-off,” said Jamil Shinawi, founder and group chief executive of AHOY. “Outperforming the biggest names in technology, on both accuracy and precision, shows that regional organisations can keep full control of their data without settling for second-best AI.”

The company said its platform can be deployed entirely within a customer’s own infrastructure and used offline, enabling organisations to keep sensitive information under their control. AHOY said internal tests indicated that the platform achieved the same benchmark performance in fully offline settings.

AHOY said these features are particularly relevant to government, defence, healthcare and energy organisations, where data-sovereignty requirements can restrict the use of public cloud services. The company said it was the highest-ranked sovereign AI system on the benchmark leaderboard.

EnterpriseRAG-Bench was created by Onyx and released as an open-source benchmark together with a research paper. Results are reviewed before they are added to the public leaderboard.