Archive · 23 Jan 2026
Pokémon Games Test AI's Intelligence
Scientists are using the old Pokémon games to test how smart AI models are.
These games help them see how well the AI can think, make decisions, and work towards goals.
Different AI models like Claude, GPT, and Gemini are playing Pokémon on a platform called Twitch.
People watch these games to see how the AI improves over time.
The games are challenging because they involve puzzles, mazes, and battles.
This helps scientists understand how to make AI better at real-world tasks.
Some AI models have even beaten the original Pokémon game, and now they are trying newer versions.
AI models like Claude, GPT, and Gemini are playing Pokémon Blue to evaluate their capabilities.
The games help track AI's reasoning, decision-making, and progress toward goals.
Streams of AI playing Pokémon on Twitch have gained significant attention and comments.
Pokémon is considered a challenging and less constrained game for AI compared to others.
AI models are now tackling Pokémon sequel games after beating the original.
- Who
- AI models Claude, GPT, and Gemini
- What
- Playing Pokémon Blue to evaluate AI capabilities
- When
- Ongoing, with streams on Twitch
- Where
- On the Twitch platform and at industry conferences
- Why
- To track AI progress and abilities in complex, time-consuming tasks
- Game
- Pokémon Blue
- AI Models
- Claude, GPT, Gemini
- Platform
- Twitch
- Key Figures
- David Hershey, Graham Neubig, Joel Zhang, Jonathan Verron
- Significance
- Evaluates AI's reasoning, decision-making, and progress toward goals
- Current Progress
- Claude Opus 4.5 is currently playing live on Twitch
- Developers
- Independent developers and AI labs