Sunnyvale, CA — Meta has teamed with Cerebras on AI inference in Meta’s new Llama API, combining Meta’s open-source Llama fashions with inference expertise from Cerebras.
Builders constructing on the Llama 4 Cerebras mannequin within the API can count on speeds as much as 18 occasions quicker than conventional GPU-based options, in line with Cerebras. “This acceleration unlocks a wholly new technology of functions which are unattainable to construct on different expertise. Conversational low latency voice, interactive code technology, immediate multi-step reasoning, and real-time brokers — all of which require chaining a number of LLM calls — can now be accomplished in seconds relatively than minutes,” Cerebras mentioned.
By partnering with Meta to serve Llama fashions from Meta’s new API service, Cerebras features publicity to an expanded developer viewers and deepens its enterprise and partnership with Meta and their unimaginable groups.
Since launching its inference options in 2024, Cerebras has delivered the world’s quickest Llama inference, serving billions of tokens by its personal AI infrastructure. The broad developer group now has direct entry to a strong, OpenAI-class various for constructing clever, real-time programs — backed by Cerebras velocity and scale.
“Cerebras is proud to make Llama API the quickest inference API on the planet,” mentioned Andrew Feldman, CEO and co-founder of Cerebras. “Builders constructing agentic and real-time apps want velocity. With Cerebras on Llama API, they will construct AI programs which are basically out of attain for main GPU-based inference clouds.”
Cerebras is the quickest AI inference answer as measured by third occasion benchmarking website Synthetic Evaluation, reaching over 2,600 token/s for Llama 4 Scout in comparison with ChatGPT at ~130 tokens/sec and DeepSeek at ~25 tokens/sec.
Builders will be capable of entry to the quickest Llama 4 inference by deciding on Cerebras from the mannequin choices throughout the Llama API. This streamlined expertise will make it straightforward to prototype, construct, and scale real-time AI functions. To join early entry to the Llama API and to expertise Cerebras velocity at this time, go to www.cerebras.ai/inference.
