Cerebras Launches World’s Fastest Inference for Meta Llama 4

Over 2,600 tokens per second performance enables instant AI interactions, real-time reasoning, blazing-fast code generation, and the next generation of agentic AI applications

SUNNYVALE, Calif.--(BUSINESS WIRE)--Cerebras Systems, the pioneer in accelerating generative AI, today announced the launch of Llama 4 inference across its CS-3 systems and on-demand cloud platform. Cerebras achieves over 2,600 tokens per second on Llama 4 Scout – 19x faster than the fastest GPU solutions as verified by Artificial Analysis, a third-party AI benchmarking service. Llama 4 is available today on Cerebras via API access powered by our U.S.-based data centers, and via Hugging Face.

“Artificial Analysis has independently benchmarked Cerebras as achieving record speeds of over 2,600 output tokens per second on Llama 4 Scout, the highest we have measured across all providers. Cerebras also delivered the fastest end-to-end response time at just 0.5 seconds in our benchmarks,” said George Cameron, Co-Founder and Chief Product Officer at Artificial Analysis. “These speeds support AI developers to build agentic multi-step applications and real-time experiences that might be significantly limited on slower systems.”

“Llama 4 on Cerebras is now available on Hugging Face, giving developers access to faster inference for Meta’s latest open models,” said Clement Delangue, CEO of Hugging Face. “We’re thrilled to expand our partnership and bring this capability to millions of developers building the future of open AI.”

World’s Fastest Llama 4 Inference

Cerebras is widely recognized as the world’s fastest AI inference provider, achieving a record 2,500 tokens per second in Llama 3.3 70B. With this launch, Cerebras extends its performance leadership with the new Llama 4 family.

“Cerebras delivers the fastest Llama 4 Scout performance in the world,” said Andrew Feldman, CEO and co-founder of Cerebras. “We’ve broken all records delivering 2,600 tokens per second while the leading GPU solution delivers 137 tokens per second. Being 19 times faster than the competition opens up AI applications in real-time reasoning, agentic workflows that GPU-based platforms just can’t match.”

Instant AI, Instant Reasoning

Cerebras inference architecture stores all model parameters entirely in on-chip SRAM, delivering memory bandwidth far beyond traditional systems. This eliminates memory transfer bottlenecks and enables ultra-fast response times. The result is a breakthrough experience: instant AI, instant reasoning, and fast code generation—all at full model fidelity.

Llama 4

Llama 4 is Meta’s most advanced open-weight model family yet, introducing a Mixture of Experts (MoE) architecture that activates only a subset of parameters per token—dramatically improving efficiency while enhancing performance. The lineup includes Llama 4 Scout, with 17B active parameters and long-context capabilities, delivering state-of-the-art performance for its size, and Llama 4 Maverick, a 400B-parameter model with 128 experts that excels at multilingual understanding, creative writing, and general assistant tasks. These models offer developers powerful tools for building sophisticated, high-quality AI applications.

“Llama 4 is the most powerful Llama model to date, with better accuracy, improved multilingual capabilities, enhanced safety, and more,” said Karl Freund, founder and principal analyst at Cambrian AI. “By delivering over 2,600 tokens per second for Scout – more than 38 times faster than closed models from OpenAI and Anthropic – Cerebras is retaining the inference performance crown as the world’s fastest Llama provider. That means Meta and Cerebras are helping developers everywhere to move faster, go deeper, and build better than ever before, including personalized multimodal experiences.”

Availability

Meta has released two models in the Llama 4 family so far—the 109-billion parameter Llama 4 Scout and the 400-billion parameter Llama 4 Maverick. The Llama 4 Scout model is available immediately on Cerebras Cloud, Cerebras CS-3 hardware, and via Hugging Face. The Llama 4 Maverick model is coming soon.

About Cerebras Systems

Cerebras Systems is a team of pioneering computer architects, computer scientists, deep learning researchers, and engineers of all types. We have come together to accelerate generative AI by building from the ground up a new class of AI supercomputer. Our flagship product, the CS-3 system, is powered by the world’s largest and fastest commercially available AI processor, our Wafer-Scale Engine-3. CS-3s are quickly and easily clustered together to make the largest AI supercomputers in the world, and make placing models on the supercomputers dead simple by avoiding the complexity of distributed computing. Cerebras Inference delivers breakthrough inference speeds, empowering customers to create cutting-edge AI applications. Leading corporations, research institutions, and governments use Cerebras solutions for the development of pathbreaking proprietary models, and to train open-source models with millions of downloads. Cerebras solutions are available through the Cerebras Cloud and on-premises. For further information, visit cerebras.ai or follow us on LinkedIn or X.