Clarifai’s New Reasoning Engine Makes AI Models Faster and Less Expensive

On Thursday, AI platform Clarifai announced a new reasoning engine that promises to make running AI models twice as fast and 40% less expensive. Designed to be adaptable to various models and cloud hosts, the system utilises a range of optimisations to maximise inference power from the same hardware.

“It’s a variety of different types of optimizations, all the way down to CUDA kernels to advanced speculative decoding techniques,” said CEO Matthew Zeiler. “You can get more out of the same cards, basically.”
The results were validated by a series of benchmark tests conducted by third-party firm Artificial Analysis, which recorded industry-best records for both throughput and latency.

The process focuses specifically on inference— the computing demands of operating an AI model that has already been trained. This demand has intensified with the rise of agentic and reasoning models, which require multiple steps in response to a single command.

Clarifai, which initially launched as a computer vision service, has increasingly focused on compute orchestration as the AI boom has drastically ramped up demand for GPUs and the data centres that house them. The company first introduced its compute platform at AWS re: Invent in December, but the new reasoning engine is the first product specifically tailored for multi-step, agent-based models.

This product launch comes amid growing pressure on AI infrastructure, which has led to a series of billion-dollar deals. OpenAI, for instance, has outlined plans for up to $1 trillion in new data centre spending, anticipating near-limitless future demand for compute power. Despite the massive hardware buildout, Clarifai’s CEO believes there is still more to be done in optimizing existing infrastructure.

“There’s software tricks that take a good model like this further, like the Clarifai reasoning engine,” Zeiler says, “but there’s also algorithm improvements that can help combat the need for gigawatt data centers. And I don’t think we’re at the end of the algorithm innovations.”