Benchmark RadarAI BENCHMARK PROFILE
R3-Bench
Evaluates resource-rational reasoning by measuring how models allocate a shared budget across six-problem suites in mathematics, competitive programming, and abstract reasoning, with scoring based on completed problems.
- Released
- 2026-08-17
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Addresses the gap between demonstrated single-problem competence and shared-budget allocation, providing a fixed dataset and protocol for studying resource allocation in LLMs.
Motivation
In cognitive science, resource rationality asks how an agent should allocate limited computation to maximize expected value.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.