AI Models Defeated Minecraft Ender Dragon
The Jev and Astra integration completed the game challenge in under nine minutes for less than one dollar.
Updated on Sept. 21, 2026 in Artificial Intelligence
Live Poll
Do you believe using specialized AI tag teams is the most efficient way to advance technology?
Two artificial intelligence models, Jev and Astra, successfully defeated the Ender Dragon in Minecraft in just 8 minutes and 43 seconds. This efficient performance utilized a System 1 architecture to drastically improve upon previous attempts.
Why it matters
The successful run demonstrates that specialized System 1 models can handle repetitive, action-oriented tasks far more rapidly and affordably than traditional System 2 language models. This approach could significantly reduce the compute resources required for autonomous agent tasks.
The operation was completed for less than $1 in token costs. Jev receives unstructured state information to execute actions while Astra manages objective planning.
The players
Jev
This AI model was developed by TypeSafe AI to execute real-time in-game actions.
Astra
This model functions as an AI planner that sets long-term objectives and travel waypoints.
TypeSafe AI
This organization is the developer behind the Jev model and its integration architecture.
Diogo Almeida
He is the founder of TypeSafe AI who led the development of the project.
Ronak Malde
He conducted the technical Minecraft test to validate the performance of the AI models.
The details
Astra functioned as the planner for waypoints, while Jev handled real-time WASD and mouse controls. By selecting from predefined options rather than generating free-form text, Jev achieved rapid, low-cost execution within the game environment.
Timeline
The Ender Dragon was defeated in September 2026.
Astra's failed 141-hour run gained recognition in September 2026.
The Tech Race
This achievement marks a departure from the reliance on complex, high-latency System 2 models for real-time task execution. It highlights a growing trend toward specialized, low-latency agents that prioritize speed and efficiency in interactive digital environments.
For developers, the open-source code available on GitHub allows for the replication and adaptation of this dual-model architecture. This approach could lead to more affordable and responsive AI agents for various automated software tasks.
The takeaway
This successful Minecraft run highlights the potential of dividing complex agent tasks into planning and execution layers. Users looking to optimize AI performance for repetitive digital tasks should consider adopting similar System 1 architectures.
Further reading
Learn more about the latest developments in Artificial Intelligence.
Source note: This article includes information reported by MakeUseOf.
Live Poll
Do you believe using specialized AI tag teams is the most efficient way to advance technology?
