Environmental Dashboard
Want to know what this is? Check here
Cumulative Environmental Report
- Total CO₂e: 6.50 kg
Estimated range: 5.46 – 7.64 kg. Equivalent to driving an average car for 26.0 km.
- Energy Consumed: 22.95 kWh
Estimated range: 20.23 – 25.89 kWh. Enough to fully charge a smartphone 2,295 times.
- Water Consumed: 68.78 Liters
Estimated range: 57.88 – 80.59 Liters. About 138 standard 0.5L water bottles.
Tracking since: September 14, 2025 | Last updated: August 23, 2026 at 11:50 PM
LLM Usage Details
- Energy Consumed: 16.8949 kWh (14.1819 – 19.8341)
- Water Consumed: 67.5701 Liters (56.6724 – 79.3759)
- CO₂e: 6.5049 kg (5.4607 – 7.6361)
- Tokens processed: 122,197,683 total — 120,440,169 input (98.6%), 1,757,514 output
Tracking since: September 28, 2025
VPS Hosting Details
- Energy Consumed: 6.0523 kWh
- Water Consumed: 1.2106 Liters
- Overall Average CPU Usage: 2.84 %
Tracking since: September 14, 2025
Note: Running on 100% renewable and carbon-free energy
Methodology & Sources
- Per-model factors: energy, carbon and water per output token come from EcoLogits, an open-source framework for estimating the impact of LLM inference. It publishes a range per model; this page uses the midpoint. Each factor is stored with its source and method in the reference database.
- Input vs. output: reading a token (prefill) is compute-bound and far cheaper than generating one (decode), which is memory-bandwidth-bound. Input tokens are counted as 1/R of an output token, with R = 20 as the central value and a declared band of 16–26. This assumption is the largest single source of uncertainty on this page.
- Models without published data: xAI publishes no figures for Grok, and some older models are absent from EcoLogits. Those use a same-class model as a proxy, flagged as such in the reference database.
- VPS: measured CPU usage against the provider's declared figures. The provider states 100% renewable energy, hence 0 kg CO₂e.
Sources: EcoLogits · TokenPowerBench (AAAI) · A Systematic Characterization of LLM Inference on GPUs (arXiv)
Notes & Disclosures
- Scope: The figures reflect estimates for the LLM inference phase and the hosting VPS energy consumption.
- These are estimates, not measurements. The central figure is a midpoint; the declared range is the honest span. Nobody outside the providers can measure this precisely, and the providers do not publish it.
- Work in Progress: This report does not yet account for other impacts, such as third-party services (Apify), network infrastructure, or the hardware lifecycle.
Methodology update, August 2, 2026: environmental factors that had no documented source were replaced with EcoLogits estimates, and the treatment of input tokens was corrected. Figures published before this date are not comparable with the current ones.