OpenAI, Cerebras launch GPT-5.6 Sol Ultrafast: 750 tokens/sec, 7× faster inference

OpenAI and Cerebras announced Ultrafast Mode on the OpenAI API, a new inference tier powered by Cerebras' Wafer-Scale Engine architecture. The service delivers up to 750 output tokens per second for GPT-5.6 Sol while maintaining output quality. Initially available to a select group of customers, access will expand as capacity grows.

The core problem Ultrafast addresses is the longstanding tradeoff between model intelligence and response speed. As models scale in capability, they incur higher computational and data movement costs, slowing inference times. Users traditionally faced a choice: wait for high-quality results or accept inferior outputs delivered quickly. Ultrafast resolves this by bringing frontier-grade intelligence to applications where latency matters.

Benchmark results demonstrate the speed advantage concretely. On the Humanity's Last Exam—a challenging benchmark of 2,500 questions typically answerable only by PhD-holders in fields such as chemistry, economics, and literature—GPT-5.6 Sol Ultrafast answered all questions in 11 hours and 11 minutes. Claude Fable 5 required 78 hours and 27 minutes for the same task, meaning Ultrafast worked through the benchmark nearly 7× faster with comparable accuracy. Cerebras performed its testing on July 10 using GPT-5.6 Sol Ultrafast with Codex on xhigh reasoning; Claude Fable 5 was tested on July 13–15 with Claude Code on xhigh reasoning. On GDP-Val, a benchmark for economically valuable knowledge work tasks, Ultrafast delivered a 5.6x end-to-end speedup with no quality degradation, tested by Cerebras on July 31, 2026.

Rohan Varma, Product at OpenAI, stated: "With GPT-5.6 Sol Ultrafast, Cerebras enables AI that keeps up with how you think, code, and collaborate. We're excited to see how workflows and applications are transformed by Ultrafast inference." Jeffrey Wang, an OpenAI researcher, noted the practical impact: "Whereas formerly I might have to wait a couple minutes for a task to finish, it now finishes for me before I even have the opportunity to context-switch. It makes me way more productive."

The speed improvement is enabled by Cerebras' Wafer-Scale Engine architecture, designed specifically for frontier AI workloads. Traditional GPU-based inference is bottlenecked by memory bandwidth: model weights must be repeatedly transferred between on-chip memory and off-chip storage to generate successive tokens. Cerebras takes a contrarian approach by packing 44 GB of SRAM on each wafer-sized chip. Weights stay on-chip, and tokens flow uninterrupted through model layers pipelined across wafers. This design scales smoothly with model size, potentially sustaining the speed advantage on future frontier models.

Concretely, GPT-5.6 Sol Ultrafast enables new modes of work. In production incident response, teams can leverage Ultrafast to root-cause and address outages quickly, preserving customer trust and preventing lost revenue. Security teams can use it to detect and respond to cyberattacks in real time. More broadly, Ultrafast enables agents to be deployed on the critical path of problems where every second counts, allowing researchers and engineers to focus their attention on deep, high-value problems while offloading parallelizable commodity tasks to Standard processing. GPT-5.6 Sol is positioned as OpenAI's best model for legal briefs, financial models, and engineering reports.