Scientific Paper
Greenpixie's AI Token Methodology: Assessing the Energy, Water and CO₂-eq Impact of AI Tokens for Open and Closed Weight Models
Joshua Horswill; Ross Hunter; Matt Clifford; James Hall · 2026-09-27
What it covers
Proposes per-token energy, water and CO₂-eq estimates using open-model GPU measurements and modeled closed-model assumptions, regression and Monte Carlo uncertainty. Submitted v1 Sep 27; v2 Sep 29, 2026.
Who it helps
Intended audience has not been documented.
Limitations
Source-specific limitations have not been reviewed. Inclusion and topic labels do not establish effectiveness or reliability.
Show citation
Joshua Horswill; Ross Hunter; Matt Clifford; James Hall. 2026-09-27. Greenpixie's AI Token Methodology: Assessing the Energy, Water and CO₂-eq Impact of AI Tokens for Open and Closed Weight Models. https://arxiv.org/abs/2609.33965v1 (AI & Environment Resource Hub; record paper-arxiv-2609-33965; collection snapshot 2026-10-04).
Classification and context
Included in the original Scientific Paper collection. Labels below are metadata based suggestions. They are not verified findings, claims of effectiveness, or endorsements.
Definitions, inclusion guidance, and examples →
Available source metadata
- Author or creator
- Joshua Horswill; Ross Hunter; Matt Clifford; James Hall
- Publisher
- arXiv preprint
- Publication date
- 2026-09-27
- Date precision
- day
- Geographic scope
- Unknown
- Language
- unknown
- Peer review
- unknown
- Source access
- Not checked
- Metadata confidence
- Metadata unconfirmed
- Last verified
- Not yet verified
- Snapshot import
- 2026-10-04
- Legacy domain
- Sustainable Computing & AI Footprint
- Legacy subdomain
- AI inference footprint measurement
- journal
- arXiv preprint
- source evidence
- https://arxiv.org/abs/2609.33965
- metadata reviewed at
- 2026-10-04
Provenance and review
- Source access
- Not checked
- Metadata confidence
- Metadata unconfirmed
- Editorial relevance
- Relevance has not been separately reviewed.
- Evidence quality
- Evidence quality has not been separately assessed.
Source sheet: Scientific Paper · Row 529 · Original ID: paper-arxiv-2609-33965.
View source spreadsheet ↗
A paper listing is not a quality assessment. Peer review and findings require source-level confirmation. Source link has not yet been checked. The import date is not the original date added.
Suggest a correction for this record →Newsletter appearances
Publication, placement, and evidence are tracked separately. A legacy coverage flag alone does not prove an appearance.
Related through shared environmental topics
Ordered by the number of shared provisional topic labels. This indicates a browsing connection, not agreement between sources.
Shared topics: Energy and electricity; Climate and greenhouse gases; Water
Scientific Paper · 2026-09-15
This article proposes a five-step state policy roadmap for data center development: conduct environmental and community impact studies, use the findings in siting decisions, reassess incentives, require large facilities to pair with off-grid clean energy and storage, and mandate community-benefit sharing. It focuses on protecting grid reliability, electricity affordability, environmental integrity, and host communities.
Source: SSRNAutomated access restricted · Metadata unconfirmed
Shared topics: Research tools and geospatial methods; Energy and electricity; Climate and greenhouse gases
Tool
Sustainable AI Group's public dashboard compares modeled AI inference energy and CO₂-eq by model, provider and session, including data center overhead assumptions. Water estimates and API are marked coming soon.
Source: ToolNot checked · Metadata unconfirmed
Shared topics: Research tools and geospatial methods; Energy and electricity; Climate and greenhouse gases
Scientific Paper · 2026-09-02
The study compares performance, energy use, processing time, and carbon emissions across dictionaries, trained classifiers, and self-hosted open LLMs for three text-analysis tasks.
Source: Social Science Computer ReviewAutomated access restricted · Metadata unconfirmed
Shared topics: Research tools and geospatial methods; Energy and electricity; Climate and greenhouse gases
Scientific Paper · 2026-10-16
This preprint examines computing demands and model skill in Earth system science and proposes joint reporting of predictive performance and energy use, including for machine learning weather models.
Source: arXiv (preprint)Not checked · Metadata unconfirmed
