Reinsurance Group of America (RGA) has released its third-quarter 2025 insights on generative AI (GenAI), outlining developments in model performance, platform integration, and global infrastructure investment.
Recent technical advances have focused on expanding model capacity and improving efficiency. Meta’s Llama 4 Scout now supports context windows up to 10 million tokens, while GPT-4.1 and Gemini 2.5 each support 1 million tokens.
These upgrades allow models to process entire policy documents, contracts, or compliance manuals in a single pass, reducing the need to split content into smaller chunks.
RGA noted that re/insurers are beginning to apply these expanded capabilities to automate document analysis and support customer operations. One area seeing increased attention is the use of Mixture-of-Experts (MoE) architecture, where models selectively activate components based on task relevance.
AI systems have also grown more capable of handling multimodal tasks. Tools such as Llama 4 Maverick and OpenAI’s GPT-Image can now process text, images, video, and audio in a single workflow.
RGA’s analysis suggests this capability could support workflows such as reviewing scanned claims documents, validating ID records, or generating explanatory visuals for customer-facing content – all within unified AI systems.
Autonomous AI agents are also gaining traction. OpenAI’s Codex-1 now operates within code repositories, while Gemini’s Agent Mode can initiate calendar tasks. Claude 4 supports end-to-end workflows with limited user intervention.
RGA highlights that such autonomous tools may be integrated into re/insurance operations to support claims processing, data entry, and internal policy audits.
RGA has previously flagged enhanced reasoning capabilities in OpenAI’s o1 model as particularly relevant to underwriting and risk assessment. These improvements allow GenAI to break down complex tasks into subtasks, produce intermediate conclusions, and generate follow-up questions, improving accuracy and reducing the likelihood of hallucinations.
Global investment in GenAI infrastructure has also intensified. AWS, Nvidia, Google, and others committed significant capital this quarter to build new AI platforms. A notable example is AWS’s $5 billion agreement with Humain, a Saudi Arabian AI firm supported by the Public Investment Fund.
With the massive growth, RGA has additionally emphasized the environmental impact of GenAI model development. As large-scale systems require substantial energy and data processing power, the firm has recommended migrating to hyperscale cloud providers that run on renewable energy.
For re/insurers looking to scale GenAI responsibly, cloud architecture decisions will play a critical role in aligning with environmental, social, and governance (ESG) goals.
RGA notes that regionalized AI models are becoming increasingly relevant to re/insurers expanding into emerging markets. By integrating local language models with regulatory and cultural frameworks, firms can tailor AI-powered services in underwriting, customer support, and policy servicing.
As models continue to scale, RGA emphasizes that re/insurers should prepare for wider integration of AI into operational and strategic decision-making. The technology is no longer limited to experimental pilots; many GenAI systems are now embedded within live business environments. This includes customer interaction tools, knowledge management systems, and risk assessment platforms.
Industry projections suggest that GenAI adoption across the North American re/insurance market is expected to reach $7.3 billion by 2034. This growth reflects not only technological advancement but also increased institutional investment in transforming legacy systems, modernizing workflows, and improving data-driven decision-making across the sector.
What are your thoughts on this story? Please feel free to share your comments below.
