Methodology: score, electricity, water, CO₂

Two separate scores: overall /100 and environmental /100

Overall score (usage quality): weighted mean of 1–5 sub-scores scaled to 100 — Quality 45, Price 30, Security & commercial use 25. Missing sub-score = excluded and weights renormalized ('partial data'). Below 55% weight available: 'Score pending' (never an invented score).

Environmental score 🌿 (distinct badge): Electricity 50, Water 50, from per-generation estimates (CO₂ derives from electricity × grid intensity, so it is not counted twice). Only one of the two available = 'partial data'. These are estimates, never vendor measurements.

Why two scores: environmental impact relies on third-party estimates far more uncertain than price or quality; mixing it into the overall score hid that uncertainty. Change of 2026-10-16: previously the overall score was Quality 35, Price 25, Security 20, Electricity 10, Water 10.

Neither score reads affiliation.

Sub-scores: where each number comes from

Own data only: every tool is rated with its own data. No page borrows another tool's score (same vendor or same category). Without verified data the sub-score stays empty ('partial data').

Price (30): regular monthly price of the first paid plan, read on the official pricing page (content/pricing.json). Annual is only a billing option and promos never count. Converted to USD at a dated ECB reference rate (shown on each page). Compared with the median, recomputed at every build, of all tools in the category with a monthly price: ≤ 0.5× median = 5; ≤ 0.9× = 4; ≤ 1.25× = 3; ≤ 2× = 2; above = 1, then clamped to 2–4 so price alone never makes the score. Unconfirmed, annual-only or pay-per-use (API) price = no price sub-score.

Safety and commercial use (25): A–D grade from our data-safety review (securite-donnees.csv) and/or A–D commercial-use grade of outputs (terms of service, content/commercial-use.json), A = 5, B = 4, C = 3, D = 2; when both exist, mean rounded down.

Quality (45): only when a public ranking exists for the tool's current model — Artificial Analysis arenas (artificialanalysis.ai) or LMArena (lmarena.ai) in the matching modality (image, video, text, voice), checked 10/23/2026, source cited on each page. A multi-model platform never borrows a third-party model's score. Best rank: 1–3 = 5; 4–10 = 4; 11–20 = 3; 21–40 = 2; above = 1. No ranking = no quality sub-score.

Electricity and water (separate environmental score, 50 + 50): per-generation estimates from content/impact-estimates.json; Electricity 5 = < 1 Wh, 4 = 1–10, 3 = 10–100, 2 = 100–1,000, 1 = > 1 kWh; Water 5 = < 1 mL, 4 = 1–10, 3 = 10–100, 2 = 100–1,000, 1 = > 1 L.

Electricity, water and CO₂ (estimates)

Never presented as service measurements. Each card shows a low/high range and an 'estimate' label.

Electricity: published baselines (e.g. Luccioni et al. 2023, arXiv 2311.16863: 1.38 Wh/image SD 1.5; 11.49 Wh/image SDXL on a lab GPU; Google 2025: 0.24 Wh median Gemini text prompt, arXiv 2508.15734). Scaled by modality in content/impact-estimates.json.

Water: mL ≈ Wh × WUE (L/kWh). Google Category 2 WUE = 1.15 L/kWh (same paper). Otherwise a documented industry-default average, explicitly labelled.

CO₂: gCO₂ ≈ (Wh / 1000) × grid intensity (gCO₂/kWh). Mix: Google market-based 94 g/kWh (2024, same paper); else Ember US/world averages when the region is undisclosed.

Hosting: provider and regions from official pages (e.g. OpenAI–Microsoft Azure partnership). If unknown: 'region not disclosed, default average used'.

Primary sources

Luccioni, Jernite, Strubell — Power Hungry Processing (arXiv 2311.16863).

Google — Measuring the environmental impact of delivering AI at Google Scale (arXiv 2508.15734).

OpenAI — Continuing Microsoft partnership (Azure hosting).

Ember Electricity Data Explorer — public grid intensities.