Extract utility usage from receipts for carbon accounting
ASUENE compared Gemini and GPT-4o models on 200 sample utility receipts, extracting transaction months, consumption and costs. Its tests found Gemini more than four seconds faster on average. The article includes sample documents, outputs and benchmark charts; deployment of the tested models is not established.
Source checked: the source link is valid and the automated fact check passed. The summary is AI-assisted. This is not human verification. Check the original source before relying on its claims.
For whom
Carbon-accounting and document-processing teams at startups or larger firms; OCR evaluation skills.
Not for
The benchmark is unsuitable as proof of accuracy on unrelated document collections.
Brain Score
- Impact
- 2/2
- Evidence
- 2/2
- Reproducibility
- 1/2
- Maturity
- 0/2
- Reach
- 1/2
Five values from 0 to 2. The total has a minimum of 1. A reading aid based on the source, not an independently measured performance result. How the score works
Still current?
Outdated
Today: repeat the benchmark with supported models, since Gemini 2.0 Flash and Flash-Lite shut down on June 1, 2026.
Augments or replaces?
Mixed
Today: AI extracts utility data, while people define reporting periods and check ambiguous quantities and units.
Original source
- Type
- Blog
- Published
- Data as of
The data date records when this dataset was included in the collection. It is separate from the publication date of the source.