DeCEAT: Decoding Carbon Emissions for AI-driven Software Testing
New study reveals small language models have distinct energy profiles, with prompt design dramatically affecting emissions.
A new research paper titled 'DeCEAT: Decoding Carbon Emissions for AI-driven Software Testing' introduces the first dedicated framework for measuring the environmental impact of using small language models (SLMs) in automated software testing. Created by researchers Pragati Kumari and Novarun Deb, DeCEAT addresses a critical gap in AI sustainability analysis, which has historically focused almost exclusively on large language models while ignoring the growing use of smaller, more specialized models.
The framework systematically evaluates SLMs using the HumanEval benchmark with adaptive prompt variants based on Anthropic's template structure. It employs CodeCarbon to measure energy consumption and carbon emissions while assessing test quality through unit test coverage metrics. The research reveals that different SLMs exhibit distinct sustainability profiles—some models prioritize lower energy use and faster execution times, while others maintain higher stability or accuracy under carbon-constrained conditions. Crucially, the study demonstrates that prompt design significantly influences both environmental and performance outcomes, making sustainability a multidimensional consideration in AI testing workflows.
This work provides software engineering teams with concrete tools to evaluate the trade-offs between test quality and environmental impact. As organizations increasingly adopt AI-assisted development tools, DeCEAT offers a methodology for making informed decisions about model selection and prompt engineering that can reduce the carbon footprint of software development while maintaining testing effectiveness. The framework represents an important step toward more sustainable AI practices in the software industry.
- DeCEAT is the first framework specifically designed to measure carbon emissions from small language models (SLMs) during automated test generation
- The study found SLMs have distinct sustainability profiles—some prioritize energy efficiency while others maintain accuracy under carbon constraints
- Prompt design significantly influences both environmental impact and performance outcomes, making it a key lever for greener AI testing
Why It Matters
Provides developers with concrete tools to reduce AI's environmental footprint while maintaining software quality through smarter model and prompt choices.