News | Things Of Business

A decade of AI stack innovation now culminates in record-breaking benchmarks for model speed and latency

WASHINGTON, Sept. 25, 2025 -- Clarifai, a global leader in AI and pioneer of the full-stack AI platform, today announced the launch of Clarifai Reasoning Engine, a breakthrough in inference performance designed specifically for agentic use cases and reasoning models.

End-to-End Response Time. Lower is better.

Since 2013, Clarifai has pioneered the modern AI stack from computer vision customization to end-to-end model operations platforms. Leveraging years of experience operating frontier models in production, Clarifai's team developed the new Clarifai Reasoning Engine with optimized kernels and novel techniques that dynamically adapt to workloads, improving generation speed over time without compromising accuracy.

The launch of Clarifai Reasoning Engine follows a recent benchmarking by Artificial Analysis against top API providers for the performance of OpenAI's gpt-oss 120B model. The Clarifai Reasoning Engine recorded new industry-best records for throughput and latency on GPUs (and even bested some specialized ASIC chips from other providers).

After recently beginning evaluation by Artificial Analysis of Clarifai's hosted gpt-oss-120b model, it now achieves new records of speed, serving over 500 tokens per second with a time to first token of 0.3 seconds. In a subsequent round of tests, the Clarifai Reasoning Engine outperformed not only all GPU-based inference implementations, but also specialized non-GPU accelerators, proving for the first time that GPU performance can match—and in some cases surpass—non-GPU architectures.

These results mark a milestone in AI operations. Combined with Clarifai's industry-leading compute orchestration technology, Clarifai Reasoning Engine delivers the speed, flexibility, efficiency, and reliability that enterprises and developers need to scale intelligent applications without being locked into a single hardware vendor.

"Agentic AI and reasoning workloads burn through tokens rapidly. They require high throughput, low latency and low prices to drive viable customer use-cases," said Matthew Zeiler, Founder & Chief Executive Officer at Clarifai. "With the Clarifai Reasoning Engine, developers can unlock a new era of speed and responsiveness with inference more affordable than the best industry offerings today. We're also using our technology and AI expertise to help model builders deliver breakthrough inference performance for their custom models on standard GPUs that is now competitive with specialty non-GPU architectures for the first time ever in our industry, with agent-friendly pricing."

Clarifai's latest release is optimized for agentic AI, purpose-built to accelerate the latest reasoning models and automation tasks across industries. Note that Clarifai's performance is not one and done, but rather with adaptive performance that learns from workload behavior, the platform continually improves speed without compromising accuracy. These improvements compound over time, especially for the repetitive tasks that are common in agentic workflows. Starting today, customers can engage with Clarifai's AI experts to apply these optimizations to their own models, further enhancing performance and economics.

About Clarifai

Clarifai is a global leader in AI and the pioneer of the full-stack AI platform that helps organizations, teams, and developers build, deploy, and operationalize AI at the highest scale and availability. Clarifai's cutting-edge AI platform supports today's modern AI technologies like Large Language Models (LLMs), Large Vision Models (LVMs), Retrieval Augmented Generation (RAG), automated data labeling, high-volume production inference, and more. Founded in 2013, Clarifai is available in cloud, on-premises, or hybrid environments and has been used to build more than 1.5 million AI models, serving more than 500,000 users across 170 countries. Learn more at www.clarifai.com.

For more information or media requests, contact: pr@clarifai.com.