Cerebras said on 9 April 2025 that it had launched Meta Llama 4 inference across its CS-3 systems and on-demand cloud platform.

Artificial Analysis measured Cerebras at over 2,600 tokens per second on Llama 4 Scout, which Cerebras said is 19 times faster than the fastest GPU solutions. It also measured an end-to-end response time of 0.5 seconds.

Llama 4 is available on Cerebras through API access from US-based data centres and through Hugging Face. “We’re thrilled to expand our partnership and bring this capability to millions of developers building the future of open AI,” said Clement Delangue, CEO of Hugging Face.

Cerebras said its architecture stores all model parameters in on-chip SRAM, which removes memory-transfer bottlenecks.

Llama 4 is Meta’s open-weight model family with a Mixture of Experts architecture, which activates only a subset of parameters per token.


Share this post
The link has been copied!