Model tiers

Miners advertise which tiers they can serve based on their hardware. The AICF router picks the smallest tier that satisfies the user's request, keeping latency low and the network balanced. Run animica miner setup to see what your machine qualifies for and install the matching bundle.

tiny

cpu

Qwen/Qwen2.5-Coder-1.5B-Instruct

Commodity hardware entry tier. Suitable for laptops & cloud micro-instances.

Params: 1.5B
Min VRAM: 0 (CPU)
Min RAM: 8 GB

small

cuda

Qwen/Qwen2.5-Coder-7B-Instruct

Single-consumer-GPU tier. The everyday workhorse — most miners land here.

Params: 7B
Min VRAM: 16 GB
Min RAM: 16 GB

flagship

cuda

deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct

Animica-native flagship. MoE; high quality at moderate inference cost.

Params: 16B MoE (2.4B active)
Min VRAM: 24 GB
Min RAM: 32 GB

large

cuda

deepseek-ai/DeepSeek-Coder-V2-Instruct

Datacenter-grade. Highest quality; reserved for capable miners.

Params: 236B MoE (21B active)
Min VRAM: 80 GB
Min RAM: 128 GB

Want to run a custom tier?

Override the auto-detected eligibility with animica miner setup --tiers tiny,small. Tier names come from ai/configs/model_catalog.yaml.