CreativityBench: Evaluating Agent Creative Reasoning via Affordance-Based Tool Repurposing
:::info Stub — Full Engineering Breakdown Coming This paper was featured on Hugging Face Daily Papers on 2026-05-06 with 20 upvotes. A full breakdown with production viability rating, implementation notes, and honest limitations is being written. Subscribe to AI Letters → :::
| Authors | Cheng Qian et al. |
| Year | 2026 |
| HF Upvotes | 20 |
| arXiv | 2605.02910 |
| Download | |
| HF Page | View on Hugging Face |
Abstract
Recent advances in large language models have led to strong performance on reasoning and environment-interaction tasks, yet their ability for creative problem-solving remains underexplored. We study this capability through the lens of creative tool use, where a model repurposes available objects by reasoning about their affordances and attributes rather than relying on canonical usage. As a first step, we introduce CreativityBench, a benchmark for evaluating affordance-based creativity in LLMs. To this end, we build a large-scale affordance knowledge base (KB) with 4K entities and 150K+ affordance annotations, explicitly linking objects, parts, attributes, and actionable uses. Building on this KB, we generate 14K grounded tasks that require identifying non-obvious yet physically plausible solutions under constraints. Evaluations across 10 state-of-the-art LLMs, including closed and open-source models, show that models can often select a plausible object, but fail to identify the correct parts, their affordances, and the underlying physical mechanism needed to solve the task, leading to a significant drop in performance. Furthermore, improvements from model scaling quickly saturate, strong general reasoning does not reliably translate to creative affordance discovery, and common inference-time strategies such as Chain-of-Thought yield limited gains. These results suggest that creative tool use remains a major challenge for current models, and that CreativityBench provides a useful testbed for studying this missing dimension of intelligence, with potential implications for planning and reasoning modules in future agents.
Engineering Breakdown
Plain English
This paper introduces CreativityBench, a benchmark for testing whether large language models can creatively repurpose tools by reasoning about their physical properties rather than just using them conventionally. The researchers built a knowledge base with 4,000 entities and 150,000+ affordance annotations (linking objects to their potential uses), then generated 14,000 tasks to evaluate whether LLMs can identify non-obvious but physically plausible alternative uses for everyday objects.
Key Engineering Insight
The core technical contribution is structuring creative reasoning as an affordance-based problem: instead of testing if models memorize creative solutions, the benchmark forces them to decompose objects into parts, attributes, and actionable properties, then compose novel use cases from those primitives. This structured approach makes creativity measurable and reproducible rather than subjective.
Why It Matters for Engineers
Production AI agents (robotics, planning systems, decision-support tools) need to handle novel situations where canonical tool use fails. Testing creative reasoning explicitly helps you catch failure modes before deployment—a robot that can't think of alternative solutions when the standard tool breaks becomes useless. This benchmark gives you a concrete way to measure and improve that capability.
Research Context
Prior work evaluated LLM reasoning through fixed benchmarks (math, logic, QA), but didn't isolate creative problem-solving, which requires generating new solutions outside training distributions. CreativityBench fills that gap by grounding creativity in formal affordance theory, enabling systematic evaluation rather than anecdotal examples. This enables the field to iterate on creativity-specific architectures and training methods instead of treating it as a byproduct of general reasoning.
:::tip Subscribe Get weekly breakdowns of papers like this in AI Letters - the newsletter for engineers building production AI systems. :::