~/AI AGENTS/4b-model-mica-v0-1-crafts-minecraft-diamond-pickaxe-in-26-decisions

4B Model Mica v0.1 Crafts Minecraft Diamond Pickaxe in 26 Decisions

Developer akivet released Mica v0.1, a 4-billion parameter decision model capable of guiding a Minecraft bot from an empty inventory to crafting a diamond pickaxe in just 26 high-level steps on its first run. Operating in a live survival environment, the model averaged 108 milliseconds per decision and completed the goal in about eight minutes of game time. The project demonstrates that small, localized language models can execute long-horizon strategic planning for autonomous agents with extremely low latency. It highlights a lightweight approach where tiny decision models evaluate discrete choices rather than generating freeform text, making local agentic AI far more practical for everyday hardware. Mica receives the textual game state and candidate commands, selecting optimal choices that a Mineflayer bot executes using Mindcraft skills. While the bot was given the coordinates of nearby diamonds to skip raw exploration, Mica made smart autonomous tradeoffs, such as choosing to mine coal over burning wood planks to smelt iron efficiently.

## BACKGROUND

Minecraft is a popular open-ended sandbox game that serves as a standard benchmark for AI research, particularly for testing embodied autonomous agents. Traditional approaches often rely on massive cloud-hosted LLMs or complex code generators, whereas frameworks like Mineflayer offer APIs that allow external scripts to interface directly with Minecraft servers.

## REFERENCES

## KEYWORDS

#AI Agents#LLM#Minecraft AI#Small Language Models#Autonomous Agents

$ subscribe --daily

4B Model Mica v0.1 Crafts Minecraft Diamond Pickaxe in 26 Decisions | Daily News