ChatGPT Ultrafast mode Pushes GPT-5.6 Sol to 750 Tokens Per Second.

OpenAI is giving its GPT-5.6 Sol model a substantial speed boost. The company has introduced Ultrafast mode, an API option capable of generating as many as 750 tokens per second, which OpenAI says is up to 14 times faster than standard inference.

The company announced the feature on LinkedIn, describing it as a way to bring its most intelligent model into products and workflows where waiting for a response carries a measurable cost. Ultrafast mode is powered by Cerebras, the AI chip and systems company known for hardware designed around extremely fast model inference.

Access is limited for now. OpenAI is launching Ultrafast mode through its API to a select group of customers, rather than making it generally available inside the ChatGPT interface. More businesses are expected to gain access as capacity grows, but OpenAI has not provided a broader rollout date, eligibility requirements, or pricing.

Model speed is often presented as a benchmark, but the practical effects depend heavily on the application. A faster model can begin and complete long responses more quickly, although the 750-token figure describes maximum output throughput and does not, by itself, explain time to first token, network latency, or the time required to process a large prompt.

For interactive products, those distinctions matter. OpenAI identified several areas where Ultrafast mode could provide an advantage:

• Real-time voice and customer support, where long pauses can make an automated conversation feel broken
• Commerce, including assistants that need to retrieve information and respond while a customer is making a purchase
• Coding and design, particularly workflows that generate or revise large amounts of content
• Financial research, where systems may need to analyse documents and return detailed findings quickly
• Security response, where delays can reduce the value of a model-generated assessment

The launch follows OpenAI’s broader expansion of GPT-5.6 Sol and GPT-5.6 Luna access. Ultrafast mode does not appear to introduce a separate model or new intelligence capability. Instead, it changes how quickly GPT-5.6 Sol can deliver its output.

That makes the update most relevant to builders whose products already depend on frontier-model performance but have found response time restrictive. A background automation that drafts a weekly report may see little benefit. A voice agent, live coding interface, support assistant, or security tool could be more sensitive to every additional second.

For small businesses and NoCode builders, throughput alone should not determine whether Ultrafast mode belongs in a production workflow. Teams will need to compare its eventual cost, access limits, reliability, and end-to-end latency with their existing API configuration. None of those operational details were included in the announcement.

It will also be important to test the complete workflow rather than an isolated prompt. External API calls, retrieval systems, databases, automation steps, and user-interface rendering can remain bottlenecks even when the model generates at hundreds of tokens per second. In some applications, shortening prompts or reducing the number of sequential model calls may produce a larger improvement than faster inference.

Still, the announcement points to a notable shift in how OpenAI is packaging model performance. Model quality has traditionally been the headline metric, while developers accepted slower responses from more capable systems. Ultrafast mode attempts to reduce that trade-off by pairing GPT-5.6 Sol with specialised Cerebras infrastructure.

The immediate constraint is availability. Until OpenAI expands access and publishes commercial terms, most developers can treat Ultrafast mode as an early look at a faster deployment tier, not a default option they can build around today.

What is ChatGPT Ultrafast mode?

ChatGPT Ultrafast mode is a high-speed inference option for GPT-5.6 Sol, powered by Cerebras. OpenAI says it can generate up to 750 tokens per second and run as much as 14 times faster than standard inference.

Who can access Ultrafast mode?

OpenAI is initially offering it through the OpenAI API to a select group of customers. The company plans to expand access to more businesses as capacity grows, but it has not announced general availability in ChatGPT.

How much does Ultrafast mode cost?

OpenAI has not disclosed pricing for Ultrafast mode. The announcement also did not specify usage limits, qualifying API plans, or a date for wider access.