OpenAI's GPT-5.6 Sol Now Runs 14x Faster With New Ultrafast Mode
News Summary
OpenAI on August 13, 2026 (Pacific Time) unveiled a preview of Ultrafast, a new inference tier for its GPT-5.6 Sol model that runs up to 14 times faster than the model's Standard processing mode, reaching output speeds of up to 750 tokens per second. The new tier is powered by hardware from Cerebras Systems and is being rolled out first through the OpenAI API to a limited group of customers, with broader access expected as capacity increases.
What Ultrafast Changes
Historically, developers who needed near-instant responses from a large language model had to trade down to a smaller or more narrowly specialized model, sacrificing reasoning quality for speed. OpenAI says Ultrafast breaks that trade-off: GPT-5.6 Sol running in Ultrafast mode retains the same underlying intelligence as GPT-5.6 Sol Standard, while completing generation dramatically faster. OpenAI frames this as a shift toward "more useful work per second" rather than simply a faster chatbot experience.
The Hardware Behind the Speed
The performance gain comes from Cerebras' Wafer-Scale Engine architecture. Unlike conventional GPU-based inference, which must continuously move model weights between on-chip memory and off-chip storage, Cerebras' wafer-sized chips keep the full set of weights on-chip, in roughly 44 gigabytes of SRAM per wafer. Cerebras said in a blog post published the same day that this design removes the memory-bandwidth bottleneck that typically limits how fast frontier-scale models can generate text on standard GPU clusters. Cerebras also confirmed the partnership through a separate press release distributed via GlobeNewswire, describing Ultrafast as evidence that frontier-level intelligence and real-time speed no longer need to be mutually exclusive.
Preview Availability and Rollout
Ultrafast is currently available only in a limited preview. OpenAI has not disclosed the exact size of the initial customer group, saying only that access will expand progressively as infrastructure capacity grows. The rollout begins with the OpenAI API rather than consumer-facing ChatGPT surfaces, suggesting the initial focus is on developers building latency-sensitive applications rather than everyday chat use.
Pricing
As of this report, OpenAI has not published a separate price for Ultrafast requests. Standard GPT-5.6 Sol pricing through the API remains at 5 US dollars per million input tokens and 30 US dollars per million output tokens; OpenAI has not said whether Ultrafast will carry a premium over these rates.
Potential Use Cases
OpenAI pointed to several categories of workloads where sub-second, high-throughput generation could matter most: incident response systems that need to summarize and act on information in real time, customer service and support tools handling live conversations, financial market analysis that depends on rapid turnaround, and e-commerce applications such as live product recommendations or checkout assistance. In each case, the pitch is that faster generation allows more complex, multi-step AI workflows to run within the latency budgets that real-time applications require.
Industry Context
The announcement continues a broader industry trend of pairing frontier language models with specialized inference hardware to cut latency, following similar speed-focused partnerships between AI labs and chipmakers such as Cerebras and Groq. Coverage of the announcement from outlets including TechCrunch and 9to5Mac noted that a 14x speed multiplier at equivalent model intelligence marks one of the more significant throughput jumps disclosed by a major AI lab so far in 2026.