OpenAI and Cerebras have unveiled GPT-5.6 Sol Ultrafast, a new service tier in the OpenAI API that delivers up to 750 output tokens per second without compromising quality. The model is initially available to a select group of customers, with access expanding over time.
TL;DR
- OpenAI and Cerebras introduce GPT-5.6 Sol Ultrafast, offering unprecedented speed and intelligence for time-sensitive tasks.
- The model achieves 750 tokens per second, 11x faster than Fable 5 and 5x faster than Opus 4.8 on Fast mode.
- Ultrafast mode is powered by Cerebras' Wafer-Scale Engine architecture, designed for efficient data movement and high-speed inference.
What happened
GPT-5.6 Sol Ultrafast is a new service tier launched by OpenAI and Cerebras, initially available to a select group of customers through the OpenAI API. The model delivers up to 750 output tokens per second, resolving the tradeoff between speed and intelligence in AI models. Cerebras' Wafer-Scale Engine architecture powers the Ultrafast mode, enabling high-speed inference by keeping model weights on-chip and minimizing data movement.
Why it matters
The introduction of GPT-5.6 Sol Ultrafast has significant implications for AI developers, startups, and investors. For developers, the model offers a powerful tool for accelerating time-sensitive, mission-critical work, such as legal briefs, financial models, and engineering reports. Startups can leverage Ultrafast to enhance their AI-powered products and services, gaining a competitive edge in the market. Investors may see this as a strategic move by OpenAI and Cerebras to dominate the AI inference market, with potential implications for the valuation and growth prospects of both companies.
Key facts
- GPT-5.6 Sol Ultrafast delivers up to 750 output tokens per second.
- The model is 11x faster than Fable 5 and 5x faster than Opus 4.8 on Fast mode.
- Ultrafast mode is powered by Cerebras' Wafer-Scale Engine architecture.
- The model achieved 100% accuracy on Humanity's Last Exam in 11 hours and 11 minutes, compared to Fable 5's 78 hours and 27 minutes.
- Ultrafast mode is initially available to a select group of customers, with access expanding over time.
- The model is designed for high-stakes work, such as root-causing production outages and responding to cyberattacks.
- Cerebras' Wafer-Scale Engine architecture packs 44 GB of SRAM on each wafer-sized chip, eliminating inefficient data movement.
Context
The launch of GPT-5.6 Sol Ultrafast comes at a time when the demand for high-speed, intelligent AI models is on the rise. As AI models continue to scale up in size and intelligence, the need for faster inference capabilities becomes increasingly important. OpenAI and Cerebras' collaboration on this project highlights the growing trend of partnerships between AI companies to drive innovation and push the boundaries of what's possible in the field of AI. The introduction of Ultrafast mode is a significant step forward in the evolution of AI inference, with potential implications for a wide range of industries and applications.
