Among the fastest hosts for open models (with Cerebras for Gemma-class). Re-benchmark periodically — latency rankings drift. Connections uses Gemma 4 uses Llama 3.3 70B part_of Large Language Model
