OpenAI accelerates GPT-5.6 Sol by up to 14x
Ultrafast shows that inference speed is becoming a new competitive dimension for frontier AI, but limited access and undisclosed pricing call for caution from Serbian software buyers.
What was announced
On August 13, OpenAI previewed Ultrafast, a new API service tier that runs GPT-5.6 Sol up to 14 times faster than standard processing, reaching up to 750 output tokens per second. Powered by Cerebras, it is currently available only to a select group of customers, with broader access planned as capacity grows.
Where speed changes the economics
OpenAI identifies incident response, customer support and voice systems, financial analysis, commerce, and interactive research as initial applications. TechCrunch reports that this remains a limited preview while OpenAI evaluates where the additional speed creates meaningful business value.
Hardware becomes part of the product
Ultrafast is not merely a software setting. Cerebras and OpenAI previously agreed to a multi-year, staged deployment of 750 megawatts of high-speed inference infrastructure beginning in 2026. This suggests that the choice of inference hardware will increasingly become an explicit product and procurement decision rather than an invisible vendor detail.
Virtual Arc’s position is that companies operating in Serbia should not buy Ultrafast simply because the headline number is impressive. It is relevant only where latency has a measurable cost: production outages, live customer calls, fraud checks, or other workflows in which waiting directly loses money or trust. For document drafting, internal search, and most background automation, cost per successfully completed task is likely to matter more than tokens per second. With access still restricted and no public price in the announcement, Serbian buyers should not redesign their stack around this tier yet. They should, however, prepare for model routing: reserve premium, low-latency inference for critical steps and send routine work to cheaper models. We believe this is more valuable than loyalty to any single provider. Faster frontier AI also weakens the traditional Serbian outsourcing proposition built around selling lower-cost engineering hours. Local software firms will increasingly need to compete through process knowledge, integration quality, reliability, regional context, and accountability for outcomes—not through the time a task consumes.