Previewing Ultrafast mode: GPT-5.6 Sol at up to 14X the speed
| Source: OpenAI Blog
Tags: OpenAI, GPT-5.6, Cerebras, inference speed, API, Ultrafast, AI infrastructure
OpenAI is previewing Ultrafast, an API tier running GPT-5.6 Sol at up to 750 output tokens per second — 14× standard speed — powered by Cerebras inference hardware, targeting latency-critical applications where throughput matters more than cost.
Details
OpenAI's Ultrafast tier delivers GPT-5.6 Sol at up to 750 output tokens per second using Cerebras hardware. The 14× speed increase over standard API rates opens use cases where inference latency is the primary bottleneck: real-time voice agents, live coding assistants, high-frequency document processing pipelines, and streaming transcription.\n\nThis is the first confirmed OpenAI production partnership with Cerebras, whose specialized inference chips achieve significantly higher memory bandwidth than standard GPUs — enabling the throughput figures described. If the Cerebras partnership expands to other models, it could signal a broader infrastructure shift away from pure GPU inference for high-speed tiers.\n\nPricing, rate limits, and availability timeline for Ultrafast are not disclosed in the announcement preview. The tier is positioned as a distinct offering for latency-critical workloads rather than a replacement for cost-optimized inference.