Qwen 3.8 27B on Cerebras: ~1500 tok/s on OpenPaths | OpenPaths Blog

Alibaba’s 27B dense multimodal on Cerebras wafer-scale inference is live as cerebras/qwen-3.8-27b: ~1500 tokens/sec, 128K context, $0.99/$1.49 per million tokens, and a one-shot pelican to prove it.