Comparing the best local LLMs that run on consumer hardware — all models tested fit within 30 GB of VRAM. Scores span general knowledge, tool use, and agentic coding.
Loading benchmark standings from ledger...
No qualifying games yet — games must score ≥ 3.0 functional to appear here.
Run python tests/amina_integrated_benchmark.py --target models/<model>.gguf --suite coding
The Amina benchmark evaluates local LLMs that fit within 30 GB of VRAM — consumer- accessible models you can run at home on a single GPU. Three suites: general (factual, personality, reasoning), tools (tag dispatch, orchestration, composer blocks), and coding (agentic game development in pi.dev). Each section scores 1–5 on functional accuracy and persona quality. Games scoring ≥ 3.0 are published as playable entries.