Cribl Introduces Telemetry Decision Bench and cribl-decision-2 Models

WHAT HAPPENED

Cribl launched Telemetry Decision Bench and cribl-decision-2, a family of domain-trained AI models designed for IT and security telemetry. The company's AI Research Lab developed smaller, specialized models that outperform larger general-purpose frontier models while using less compute and delivering lower latency.

WHO IS INVOLVED

Cribl built these models internally through its AI Research Lab. The announcement references competing models including Jev 1.14 and Microsoft Decisions 1 as performance benchmarks.

HOW IT WORKS

Cribl trained domain-specific models ranging from 0.8 billion to 4 billion parameters on IT and security telemetry tasks. The company evaluated performance using Telemetry Decision Bench, a custom benchmark comprising 4,800 production-like IT and security tasks. The smaller 0.8B model beat competing models in its size class by 8.4 percentage points in accuracy, while the 4B domain-trained model matched the performance of a much larger general-purpose predecessor while running 7 times faster.

WHY IT MATTERS

Specialized AI models deliver comparable accuracy to leading proprietary models while running 1.3 to 4.1 times faster, reducing latency and computational overhead for IT and security operations. This approach demonstrates that domain-specific training can outweigh raw model size, making AI more practical for organizations managing telemetry data at scale.

THE DEAL

Cribl released Telemetry Decision Bench publicly with a detailed benchmark report and additional documentation available online.

Read more

• linkedin.com ↗