Models in the Umbra catalog
Live registry from the coordinator. Each entry is a public HF repo, pinned to a known-good revision and GGUF digest, priced per million tokens. Click any model to use it in the playground.
arch = model family (llama / qwen / mixtral / etc.)quant = GGUF precision (Q4_K_M = ~4 bits/weight)min_unified_memory_gb = M-series RAM floor to load