April 29, 2025 – Sunnyvale, CA – Meta has partnered with Cerebras to introduce ultra-fast inference in its new Llama API. This collaboration brings together the widely popular open-source Llama models with the world's most advanced inference technology, setting the stage for transformative applications for developers worldwide. The integration of Cerebras' technology with the Llama API promises to deliver generation speeds up to 18 times faster than traditional GPU-based solutions. This unprecedented acceleration opens the door to a new generation of applications that were previously unfeasible. Real-time conversational voice, interactive code generation, instant multi-step reasoning, and dynamic real-time agents can now be developed with ease, completing tasks in seconds instead of minutes. By serving Llama models from Meta’s API service, Cerebras enhances its visibility and expands its reach to a broader global developer audience. This move also solidifies Cerebras’ existing relationship with Meta and its talented teams. Since launching its inference solutions in 2024, Cerebras has been recognized for delivering the world's fastest Llama inference, processing billions of tokens through its cutting-edge AI infrastructure. The community now has access to a powerful, OpenAI-grade alternative for building intelligent, real-time systems, backed by Cerebras’ unparalleled speed and scale. Andrew Feldman, CEO and co-founder of Cerebras, expressed his enthusiasm about the collaboration: "Cerebras is proud to make the Llama API the fastest inference API in the world. Developers need speed to build agentic and real-time applications. With Cerebras on Llama API, they can create AI systems that are beyond the capabilities of leading GPU-based inference platforms." Artificial Analysis, a third-party benchmarking site, confirms Cerebras' leadership in AI inference performance. The site ranks Cerebras as the fastest solution, achieving over 2,600 tokens per second for Llama 4 Scout, compared to around 130 tokens per second for ChatGPT and approximately 25 tokens per second for DeepSeek. To access the fastest Llama 4 inference, developers can simply select Cerebras from the model options within the Llama API. This user-friendly approach simplifies the process of prototyping, building, and scaling real-time AI applications. Early access to the Llama API and the Cerebras speed advantage is now available at www.cerebras.ai/inference. About Cerebras Systems Cerebras Systems is a renowned team of innovative computer architects, computer scientists, deep learning researchers, and engineers. Their mission is to revolutionize generative AI by building a new class of AI supercomputers from the ground up. The company’s flagship product, the CS-3 system, is powered by the world’s largest and fastest commercially available AI processor, the Wafer-Scale Engine-3. These systems can be quickly and easily clustered to form the most powerful AI supercomputers globally, simplifying model deployment by eliminating the complexities of distributed computing. Cerebras Inference offers groundbreaking speeds that enable customers to develop state-of-the-art AI applications. Prominent corporations, research institutions, and government entities utilize Cerebras solutions for developing proprietary models and training open-source models, which have been downloaded millions of times. Cerebras solutions are available both through the Cerebras Cloud and on-premises installations. For more information, visit cerebras.ai or connect with Cerebras on LinkedIn, X, and Threads. Media Contact [email protected]