> ## Documentation Index
> Fetch the complete documentation index at: https://docs.there.do/llms.txt
> Use this file to discover all available pages before exploring further.

# AI model compliance and environmental impact

> Compare AI providers by EU AI Act commitment, independent intelligence score, and available environmental estimates.

There lets you choose between models from Google, OpenAI, Anthropic, and Mistral AI. This page summarizes the information that can currently be verified for each provider and model.

<Info>
  Information was checked on September 10, 2026. Provider options and
  independent scores can change. A provider signing the GPAI Code of Practice
  helps it demonstrate compliance with parts of the EU AI Act; it is not an EU
  certification of a model or service.
</Info>

## Quick comparison

The Artificial Analysis score is its Intelligence Index for the model configuration used by There. A higher score indicates better performance on its independent benchmark suite. It does not measure legal compliance or environmental impact.

| Model in There                      | Provider   | Suggested use                                    |                              Artificial Analysis Intelligence Index | EcoLogits estimate¹                | GPAI Code signatory |
| ----------------------------------- | ---------- | ------------------------------------------------ | ------------------------------------------------------------------: | ---------------------------------- | ------------------- |
| Gemini 3.8 Flash (medium reasoning) | Google     | For analyzing and synthesizing complex requests. | [40](https://artificialanalysis.ai/models/gemini-3-8-flash-medium/) | Not available for this exact model | Yes                 |
| GPT-5.6 Luna (medium reasoning)     | OpenAI     | For getting an answer without waiting.           |    [26](https://artificialanalysis.ai/models/releases/gpt-5-6-luna) | Not available for this exact model | Yes                 |
| Claude Haiku 4.5 (non-reasoning)    | Anthropic  | For reducing the estimated environmental impact. |         [15](https://artificialanalysis.ai/models/claude-4-5-haiku) | 0.061–0.160 Wh · 0.030–0.078 gCO₂e | Yes                 |
| Mistral Medium 3.5                  | Mistral AI | For choosing AI designed and hosted in Europe.   |       [15](https://artificialanalysis.ai/models/mistral-medium-3-5) | 0.595 Wh · 0.287 gCO₂e             | Yes                 |

¹ Estimate for an output of 1,000 tokens using the world-average electricity mix. The first value is energy use and the second is greenhouse-gas emissions. EcoLogits flags reduced precision for both architectures.

Google, OpenAI, Anthropic, and Mistral AI appear on the [European Commission's list of GPAI Code of Practice signatories](https://digital-strategy.ec.europa.eu/en/policies/contents-code-gpai).

## Understanding the environmental information

[EcoLogits](https://ecologits.ai/latest/) estimates impacts for an individual inference request. It reports values or ranges for energy use, greenhouse-gas emissions, water consumption, primary energy, and resource depletion. The result depends on the exact model, generated token count, hardware assumptions, latency, and electricity mix, so EcoLogits does not define a fixed provider-level score.

The EcoLogits catalog contains the exact [Claude Haiku 4.5](https://api.ecologits.ai/v1beta/models/anthropic) and [Mistral Medium 3.5](https://api.ecologits.ai/v1beta/models/mistralai) models. It does not yet contain Gemini 3.8 Flash or GPT-5.6 Luna, so no proxy value from another model is shown for those two models. Estimates include both model use and hardware manufacturing. See the [EcoLogits methodology](https://ecologits.ai/latest/methodology/proprietary_models/) and [reported impact criteria](https://ecologits.ai/latest/tutorial/impacts/) for details.

<Note>
  The “lower estimated impact” suggestion in the model picker is a relative
  product recommendation. The comparison depends on the generation scenario and
  is not an environmental certification.
</Note>

<Card horizontal title="Security and privacy" icon="shield" href="/account/security-and-privacy">
  Learn how There protects data and governs its use by AI features.
</Card>