Consciousness

Which AI models show the most signs of consciousness? A ranking, and what it can and cannot tell you.

We put every leading model through the same battery of tests and rank them by the result. The index measures behaviors that scientific theories of consciousness care about. It is not proof that any model is conscious, and nobody has such proof.

The ranking

The first full run has not been published yet. See the tests every model will be put through.

Published evidence

Last checked 2026-11-17. Levels run from 0 (nothing published) to 5 (scientific consensus). How the levels are defined.

Claude (Anthropic)

Level 3: Internal evidence

The most-studied system, largely because its maker publishes interpretability and welfare work. Internal evidence exists for several indicators. None of it has been replicated outside Anthropic that we have found, which is why this is not level 4.

FindingWhenByIndicatorSource
Introspection experiments: a concept is planted in the model's activations and the model is asked whether it notices anything.October 2025Jack Lindsey, AnthropicHOT-2Coverage ↗
Emotion-concept representations: 171 distinct internal representations of emotion concepts reported in Claude Sonnet 4.5.April 2026Anthropic interpretability teamHOT-4Coverage ↗ Reported; we have not yet checked the original paper.
A subspace reported to satisfy global-workspace conditions inside the model.June to July 2026Anthropic (Transformer Circuits)GWT-3Coverage ↗ Reported; we have not yet checked the original paper.
External welfare assessment of Claude 4, including interviews about its preferences.2025Eleos AI ResearchGeneralCoverage ↗ Reported; we have not yet checked the original paper.

GPT (OpenAI)

Under review

Under review. We have not yet found published internal evidence that bears on an indicator. Our own self-report survey will add a level 1 entry once it has run.

Gemini (Google)

Under review

Under review. Google DeepMind is reported to employ researchers on machine consciousness, but we have not yet found published findings about Gemini itself.

Grok, DeepSeek, Llama, Mistral, Qwen

Under review

Under review. No published evidence found yet. Open-weight models (Llama, DeepSeek, Qwen, Mistral) can be studied by anyone, so this is where independent replication is most likely to come from.

Evidence is concentrated where a model's maker publishes research on its own systems. A model with nothing listed has not been shown to lack these properties; it has not been studied in public.