The world's first insurance-native language model
Reinsure-8B is a small language model built exclusively for the reinsurance and insurance industry. Fine-tuned from Llama 3.1 on reinsurance workflows, treaty structures, bordereaux formats, and London Market language.
Deploy it sovereign inside your own infrastructure, or call it as an inference API. Either way, you get a model that speaks insurance — without prompt engineering, without hallucinated policy terms.
What is Organisational Sovereign AI?
Sovereign AI means the model runs inside your control. Your weights, your infrastructure, your data. No dependency on a third-party API that can change pricing, deprecate versions, or inspect your query traffic.
For regulated insurance businesses — Lloyd's syndicates, global reinsurers, captives, MGAs — sovereignty is not a preference. It is a compliance requirement. Sensitive submissions, treaty terms, and client data cannot flow to shared cloud endpoints.
Reinsure-8B is purpose-sized for sovereign deployment. At 8 billion parameters, it runs efficiently on enterprise GPU hardware — a single A100, L4, or equivalent — without the infrastructure overhead of 70B+ models. This is the practical path to production AI in a regulated industry.
Two ways to run Reinsure-8B
- Sovereign Deployment: Deploy the full model weights inside your own cloud or on-prem environment. Fine-tune it on your proprietary data. Complete data sovereignty and maximum performance.
- Hosted API: Call Reinsure-8B as a hosted API via Reinsured.AI. No infrastructure required. Pay per inference token. Ideal for teams validating use cases or building lightweight integrations.
Built for the language of reinsurance
Generic large language models are trained on the open internet — predominantly consumer content, code, and general text. Insurance knowledge is sparse, often incorrect, and never updated with current market practice.
Reinsure-8B was fine-tuned on a curated corpus of reinsurance-specific content — treaty wordings, bordereaux templates, Lloyd's market standards, catastrophe model outputs, underwriting guidelines, and claims documentation.
The result is a model that interprets reinsurance language correctly by default, without requiring you to explain what a "binder", "burning cost", or "cedant" means in every prompt.
Domain-specific beats general-purpose
Applying a general-purpose model to insurance creates compounding problems — hallucination, data risk, poor economics. Reinsure-8B is purpose-built to eliminate each of them.
The intelligence engine behind the stack
Reinsure-8B is the reasoning core that powers Reinsured.AI's Context Cloud and AI Agents. When an agent interprets a treaty clause, extracts a bordereaux field, or classifies a submission, it is calling on Reinsure-8B — a model that already understands the domain.
Organisations that deploy Reinsure-8B sovereign get the additional option to fine-tune it on their own internal data, creating a model layer unique to their underwriting philosophy and market positioning — one that becomes a proprietary asset over time.
Technical Specifications & Deployment
This model has been exported to standard 16-bit Safetensors format. It is fully compatible with industry-standard cloud inference engines, including vLLM and HuggingFace Text Generation Inference (TGI).
Cloud Native Example (vLLM / Modal)
For organizations deploying Sovereign AI at scale with scale-to-zero economics, Reinsure-8B is optimized for vLLM. Here is a reference architecture deploying to Modal Serverless:
import modal
vllm_image = modal.Image.debian_slim().pip_install("vllm")
app = modal.App("reinsure-8b-sovereign")
@app.function(
image=vllm_image,
gpu="L4", # Extremely cost-effective for 8B models
container_idle_timeout=300 # Scale to zero when inactive
)
@modal.web_server(8000)
def serve():
import sys, subprocess
subprocess.Popen([
sys.executable, "-m", "vllm.entrypoints.openai.api_server",
"--model", "Reinsured-AI/Reinsure-8B",
"--port", "8000",
"--max-model-len", "8192"
])