SambaNova Launches Fastest Llama 4 Maverick Inference Platform

SambaNova, the generative AI company delivering the most efficient AI chips and fastest models, today announced the availability of Meta’s Llama 4 Maverick model on SambaNova Cloud — setting a new industry benchmark delivering 655 tokens per second inference speed. The performance, verified by independent evaluator Artificial Analysis, makes SambaNova the fastest platform for Llama 4 Maverick inference.

“Artificial Analysis has independently benchmarked SambaNova’s cloud deployment of Meta’s Llama 4 Maverick model (400B total parameters, 17B active parameters) at 655 output tokens/s. This is the fastest output speed we have measured yet for Llama 4 Maverick, and it is several times faster than the fastest speeds achieved so far on GPUs. Llama 4 Maverick is Meta’s strongest model yet and a top choice for a wide range of workloads,” said Micah Hill-Smith, CEO & Co-Founder of Artificial Analysis.

Llama 4 Maverick, Meta’s flagship multimodal model featuring 400 billion total parameters/17 billion active parameters and 128 experts, outperforms industry competitors such as GPT-4o and Gemini 2.0 Flash in multilingual and visual understanding benchmarks. Utilizing SambaNova’s advanced Reconfigurable Dataflow Unit (RDU) chips, Maverick on SambaNova Cloud delivers industry-leading performance with unparalleled computational efficiency.

“We’re thrilled to partner with Meta, bringing the fastest inference speeds ever recorded for Llama 4 Maverick to developers and enterprises,” stated Rodrigo Liang, CEO and co-founder of SambaNova. “This collaboration marks a significant leap forward in the efficiency and capability of multimodal AI models.”

This week, SambaNova will deliver multimodal capabilities and longer context lengths with Meta’s Llama 4 Maverick, enabling developers with image recognition and agentic capabilities.