Bevaya's Insurance-Trained AI Beats General Models in Benchmark Test

NEW YORK, Sept. 23, 2026 /CourierPR/ -- Bevaya, an AI platform specialized in insurance, has demonstrated superior performance in a benchmark test against leading general-purpose AI models. The results, published on Bevaya’s website, highlight the company’s InsurGPT™ loss run model’s ability to outperform its competitors in accuracy and speed.

Across 346 real loss runs, Bevaya’s insurance-trained model achieved a field accuracy rate of 93.1%, significantly higher than the 85.3% accuracy of the strongest general-purpose model tested. Every general-purpose model, regardless of its generation or price, scored between 78% and 85%. This benchmark, conducted to measure the models’ performance on the industry’s most challenging documents, underscores the importance of specialized training for AI in insurance.

The difference in performance is attributed to the unique training process Bevaya employs. InsurGPT™ is trained on more than 300 million non-public insurance documents, meticulously labeled by insurance practitioners. This specialized training ensures that the model can accurately interpret and analyze the complex and varied formats of insurance documents, which are often formatted differently by each carrier.

Chaz Perera, co-founder and CEO of Bevaya, emphasized the significance of this training process.

The benchmark results also showed that Bevaya’s model was more efficient, with fewer errors and faster processing times. On average, the model read each document in less than half the time taken by the general-purpose models. Additionally, Bevaya’s model was more accurate in identifying critical fields such as claim numbers, policy numbers, and claimant IDs, where general-purpose models scored as low as 59%.

In production, Bevaya adds a second model to verify every answer against the source document and provides a confidence score on every field. Anything uncertain is flagged for review by the insurer’s own staff through Bevaya’s patented human-in-the-loop technology, ensuring that the final output is 98% accurate or higher.

Ratish Dalvi, SVP of AI and engineering at Bevaya, highlighted the importance of this benchmark.

Bevaya has delivered more than 120 production deployments, including at three of the top five U.S. property and casualty carriers. The company’s AI agents read, analyze, and recommend across underwriting, claims, and policy servicing, returning verified data that professionals can act on. The benchmark isolates the model behind that work and measures it against current general-purpose AI under identical conditions.

The full results, the method, and the Bevaya Labs research are available on the company’s website at www.bevaya.ai/benchmarks.