PollySign in

About Polly

Polly is an AI agent for people who negotiate environmental agreements. She explains negotiating text, helps you prepare, and lets you rehearse in a practice room.

She is operated by GainForest for the Youth Negotiators Academy, alongside the AI for COP guidebook.

She is an AI model and can be wrong. Check anything that matters against the official text.

Your environmental footprint

Every answer takes electricity and cooling water in a data centre. This is an estimate of what your use of Polly has taken: your chat, the web searches she runs for you, and your practice rooms.

Sign in to see your own figures. The method below is the same for everyone.

How it is worked out

  1. Polly counts the tokens (pieces of words) the model reads and writes for you.
  2. Writing one token takes about 2 calculations for each active parameter of the model. Polly runs on GLM-5.3 Flash, which uses 18 billion of its 320 billion parameters per token.
  3. One data-centre chip (an H100) can do 989 trillion calculations a second. While writing it reaches 17.54% of that, the middle value of 33 metered readings in the ML.ENERGY Benchmark.
  4. The chip, with its share of the server and the building, is rated at 1,500 W and draws 70% of that on average.
  5. That makes 0.0605 Wh for every 1,000 tokens written. A token read counts as 0.25 of a written one, and a token read from cache as 0.1.
  6. Water is 1.1 litres for each kWh, for cooling. Carbon is 480 g CO₂e for each kWh, close to the world average for electricity.

Not counted: training the model, the water used to make the electricity, the search engine behind a web search, your own device, the network, and making the hardware.

How sure: this is an estimate, good to about a factor of two, not a measurement. For comparison, Google measured a typical Gemini text prompt at 0.24 Wh, 0.26 mL of water and 0.03 g CO₂e in August 2025.

Days are counted in UTC. Deleting your data on the Memory page also sets these figures back to zero. The method and the numbers are the same as in the ai4cop.org Playground.

Read the guidebook chapter: The environmental footprint