OpenAI high-speed inference tier that serves model output far faster than standard, reportedly around 14x.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Ultra Fast mode is a high-speed inference option from OpenAI. It was described by an OpenAI figure (Tibo) in a Matthew Berman interview as running roughly 14x faster than standard serving. The speaker said it is used internally for high-stakes situations such as handling outages.
According to the same discussion, the gains depend on the workload. Generation-heavy tasks see about 10x improvement, while tasks dominated by tool calls see only about 3-4x, since latency outside token generation is not reduced. The speaker expects speeds of this kind to become near-default within a year or two.
No official documentation, pricing, model list or availability details for this mode appear in the provided material, so how it is accessed, what it costs and which models support it are unconfirmed. The details here come from a single creator interview and should be treated as provisional.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
A high-speed inference tier roughly 14x faster than standard. It is used internally for high-stakes cases like outages. It gives 10x gains on generation-heavy work but only 3-4x when tool calls dominate. Tibo expects such speeds to become near-default in a year or two.
Closest neighbours by meaning, not just by category.
Model · OpenAI
OpenAI API service tier running GPT-5.6 Sol up to 14x faster than Standard, at up to 750 tokens/sec on Cerebras hardware.
Model · OpenAI
OpenAI's GPT-5.6 Sol frontier model, with an Ultrafast API tier on Cerebras reaching up to 750 output tokens per second.
API · Inference.net
OpenAI-compatible inference API and gateway for open-weight and closed models, with provider failover and pass-through pricing.
Library
Open-source streaming MoE inference framework for running large sparse models on laptops, phones and single-GPU PCs.
Model · Google
Google's faster, lower-cost Flash model in the Gemini 3.5 family, aimed at agentic and coding workloads.
Model · StepFun
Open-weights multimodal flash model from StepFun for agentic tasks involving images, interfaces, web search and long runs.
