Microsoft reported that Microsoft Discovery Engine with CLIO (Cognitive Loop via In-Situ Optimization) achieved the highest scores among agentic harnesses evaluated on Agent’s Last Exam, a demanding benchmark of long-running, tool-using professional tasks across three scientific domains.

Benchmark results

  • 61.6% in health and medicine
  • 75.2% in physical sciences
  • 64.6% in life sciences

How CLIO’s adaptive reasoning works

flowchart LR
    A[Multiple independent reasoning paths explore a problem] --> B[Compare and share learning across paths]
    B --> C{Strong evidence-backed result found?}
    C -->|No| D[Change strategy, switch model, or involve a domain expert]
    D --> A
    C -->|Yes| E[Resolve into single evidence-backed result]

CLIO is designed to determine when to keep exploring, change strategy, use a different model, or bring a human domain expert into the loop — rather than stopping at a single model response.

Real-world impact

Beyond the benchmark, Discovery Engine with CLIO has already supported real R&D work, including the discovery of a novel organic redox flow battery, and is positioned for use in materials and molecular discovery, formulation and process optimization, and lab automation.