Media
2 items
Description
Carbon transparency for AI usage — see the carbon footprint of every ChatGPT, Claude, and Gemini query. InferenceCarbon gives you an insight into the carbon cost of your AI conversations. Every query to ChatGPT, Claude, or Gemini uses energy. InferenceCarbon shows you our best estimate -- based on academic research -- of how much carbon that produces, in real time. How it works: InferenceCarbon works out of the box — no account is needed. A small badge appears on Gemini/ChatGPT/Claude pages showing each query's estimated carbon. Token counts, and carbon estimates are calculated instantly using research-derived energy consumption data. Optional: create a free account on inferencecarbon.ai to sync your data across devices and get long-term analytics. The core carbon tracker works entirely locally in your browser, without any login. Supported platforms: - ChatGPT (chatgpt.com) - Claude (claude.ai) (will automatically enable when data published) - Google Gemini (gemini.google.com) (will automatically enable when data published) What you see Per-query carbon estimate (gCO2e) with confidence bands Session totals and daily/weekly summaries Optional dashboard sync for long-term tracking Open methodology Estimates use a published, documented framework: dual GHG-Protocol Scope 2 reporting (location- and market-based), explicit uncertainty ranges, and IPCC-style confidence labels. Per-model figures come from our public dataset and refresh automatically in the background. When a model's figure hasn't been published yet, the overlay honestly says "Measuring" instead of guessing a number. Why it matters AI usage is growing exponentially, but its environmental cost is invisible. InferenceCarbon makes that cost visible — helping you make informed choices about which models to use and when. We never store or transmit the content of your prompts or AI responses. The extension reads them locally, in your browser, only to count words/tokens for the carbon estimate — the content itself never leaves your device. Only aggregate metadata (token counts, model, timestamp) is transmitted. All data stays on your device or can be synced to our cloud platform for analysis. Optional anonymous telemetry is enabled by default, and you can disable it if you want with no impact on functionality. Export or delete your data at any time. Full privacy policy: https://www.inferencecarbon.ai/privacy Methodology and details: https://www.inferencecarbon.ai Updates in this version: Enhanced gCO2e location-based and market-based data in compliance with the GHG protocol. What's new in 1.0.17 Gemini • Images generated in Gemini — including Google Workspace and the newer image interfaces — are now detected and measured correctly, instead of being counted as text. • Video output (Veo) is recognised and measured as video. • Image generation no longer hangs — the estimate finishes as soon as the image appears, rather than waiting out a timeout. ChatGPT • More accurate model detection (e.g. GPT-5.5 is no longer mistaken for another model) and awareness of reasoning effort. • Fixed double-counting of reasoning on reasoning-capable models. Claude • Fixed a timing issue that could leave a query showing "measuring" indefinitely. General • The carbon overlay now appears only on models we have data for. • New models are recognised automatically, without waiting for a store update. • The Downloads permission is now optional — requested only if you export data. • Reliability fixes and interface polish (clearer buttons, a dashboard shortcut, and a smoother first run).