Clarifai and Vultr announced new benchmark results demonstrating unprecedented inference speed and cost efficiency on GPUs. The announcement comes as the two companies showcase their collaboration at the Vultr booth during NVIDIA GTC Washington, D.C., taking place October 28-30.
The joint results highlight the Clarifai Reasoning Engine, optimized for agentic AI inference, running on Vultr’s large-scale dedicated GPU clusters. Independent testing by Artificial Analysis showed the Clarifai Reasoning Engine achieving 544 tokens per second, 0.36 seconds time to first token, and an industry-leading $0.16 per million tokens on the GPT-OSS-120B model-outpacing every GPU-based inference platform tested, and rivaling specialized ASIC accelerators.
“Clarifai’s benchmark-topping performance is a testament to what’s possible when software innovation meets cloud engineering excellence,” said Kevin Cochrane, CMO of Vultr. “Our GPU clusters deliver extraordinary inference speed and efficiency while keeping costs under control. Together with Clarifai, we’re enabling organizations to develop competitive, high-performance AI solutions faster than ever before.”
“Vultr’s GPU infrastructure is a cornerstone of how we help customers unlock the full potential of the Clarifai Reasoning Engine,” said Matthew Zeiler, Founder & CEO at Clarifai. “As organizations build AI workloads and agentic systems, they need performance and cost efficiency without compromising reasoning quality. Vultr provides the speed and reliability that make those outcomes possible.”
The latest benchmarks are part of the Clarifai 11.9 release, which introduces expanded capabilities for building and deploying advanced AI systems, including:
- Added new Vultr cloud instances on NVIDIA HGX B200 and NVIDIA GH200 Grace Hopper Superchips.
- Toolkit compatibility for vLLM, LMStudio, and Hugging Face, enabling seamless model initialization and management.
- New model releases, including Qwen3-Next-80B-A3B-Thinking and Qwen3-30B-A3B-Instruct, further expanding Clarifai’s model-agnostic ecosystem.
Built for enterprise-scale workloads, the Clarifai Reasoning Engine continuously optimizes kernels, batching, and memory use based on workload behavior, improving performance over time without sacrificing accuracy. It supports any large reasoning model, allowing customers to bring their own models and achieve similar levels of throughput and efficiency.
Clarifai and Vultr are setting a new standard for AI inference performance, scalability, and cost control, enabling developers and enterprises to accelerate innovation in reasoning, agentic systems, and generative AI.
