OpenAI's Ultrafast GPT-5.6 Sol preview targets latency-sensitive workflows with Cerebras-backed speed boost

OpenAI is previewing a new API service tier called Ultrafast that runs its flagship GPT-5.6 Sol model up to 14 times faster than standard processing, generating up to 750 output tokens per second. Powered by Cerebras hardware, the tier targets builders who need frontier model intelligence without the latency penalty, but access is restricted to a select group of customers during an initial evaluation.

Ultrafast mode delivers up to 750 tokens per second via Cerebras hardware

OpenAI introduced the GPT-5.6 model family in June with three variants: Sol as the flagship, Terra as a balanced option, and Luna optimized for speed. The full family became broadly available in July across ChatGPT, Codex, and the API. Now, OpenAI is previewing Ultrafast, a premium API tier that runs Sol at dramatically higher speeds. According to OpenAI, Ultrafast can achieve up to 14x faster processing and up to 750 output tokens per second, powered by Cerebras systems. The company says the speed comes without quality compromise.

Latency-sensitive workflows are the primary target

Ultrafast is designed for workloads where latency matters as much as model capability. OpenAI lists voice, customer support, commerce, developer agents, financial research, and security response as target use cases. For builders, this means frontier-level reasoning could be applied to real-time interactions that previously required smaller, faster models. OpenAI also notes internal use by its engineers to analyze logs during incidents and compress research cycles that previously ran overnight into multiple iterations during the workday.

Access is limited to a preview waitlist

Ultrafast launches first through the OpenAI API, but access is not open to everyone. During the initial evaluation, only a select group of customers can use it. Businesses can join a waitlist by sharing their workload, latency requirements, and expected usage. There is no announced timeline for broader availability or pricing details. This means most builders cannot rely on Ultrafast for production yet, but those with latency-critical applications should apply to test its impact.

What remains unclear about Ultrafast

Several details are missing from the preview. OpenAI has not disclosed pricing for Ultrafast, how it compares to standard Sol API pricing, or whether it will be available as a separate tier or an add-on. The 14x speed claim is based on OpenAI's internal testing, and real-world performance will depend on workload characteristics, concurrency, and network latency. Additionally, Ultrafast is powered by Cerebras hardware, which may have different availability regions or capacity constraints compared to OpenAI's standard infrastructure. Builders should treat this as an early preview and plan for potential changes as OpenAI expands access.