GPT-5.6 Sol Ultrafast: Up to 14x Faster Output

GPT-5.6 Sol Ultrafast, an OpenAI service tier up to 14x faster

Ultrafast is a service tier from OpenAI that runs GPT-5.6 Sol up to 14x faster than Standard processing. OpenAI announced it on August 13, 2026, and the tier launches first in the OpenAI API. A service tier is a speed class for the same model.

Standard processing is the baseline the 14x figure is measured against. For the model itself, see GPT-5.6 Sol, Terra and Luna: Specs and Pricing.

OpenAI frames the release around a trade-off. Until now, real-time speed typically meant choosing a smaller or more specialized model. OpenAI presents Ultrafast as progress toward more useful work per second.

Up to 750 tokens per second in Ultrafast mode

OpenAI says Ultrafast generates up to 750 output tokens per second. A token is a small unit of text that a model reads and writes, so a count of tokens per second measures how fast an answer streams out. Both numbers in the announcement carry the words "up to", which makes them ceilings, not guarantees.

The 14x comparison has one reference point, Standard processing. The announcement does not compare it with any other speed option. OpenAI also ties the speed to its pitch for the tier: when speed no longer requires giving up intelligence, AI can move into the most time-sensitive parts of a business.

OpenAI's Cerebras inference partnership behind Ultrafast

OpenAI calls Ultrafast the next step in its partnership with Cerebras, aimed at bringing ultra-low-latency inference to OpenAI's platform. Inference is the step in which a trained model produces an answer, and low latency means a short wait between a request and the reply. The same announcement credits Cerebras with powering the tier that generates up to 750 output tokens per second. The announcement lines quoted below name the partner and the goal, but they do not describe how Cerebras hardware produces the speed.

Ultrafast mode use cases for GPT-5.6 Sol, from outages to checkout

OpenAI lists five uses for Ultrafast. The fifth, live research, is tied to OpenAI's own use of the tier.

• Incident response and reliability: when a critical system fails, the model analyzes application logs, recent code changes and engineer reports to identify the likely cause and help prepare a fix while the outage is still unfolding.
• Financial research and security: it analyzes market signals, assesses transactions and identifies suspicious activity while conditions are still changing.
• Customer support and voice: it resolves complex customer issues in real time without interrupting the conversation, even when finding the answer involves multiple steps or systems.
• Commerce: it answers product questions, checks inventory, personalizes recommendations and resolves checkout issues while the shopper is still deciding, before hesitation becomes an abandoned cart.

How OpenAI uses Ultrafast mode in research and live experiments

OpenAI's research team says it sees Ultrafast tightening its experiment loop to support multiple iterations during the workday. The team's common workflow is to launch a batch of experiments overnight and review the results the next morning. With the faster tier, OpenAI says that single overnight cycle is giving way to several rounds during the day.

The same idea is the fifth use case on OpenAI's list, live research and experimentation. In OpenAI's description, research that previously took an overnight run becomes an interactive working session. A team can test an idea, examine the results, adjust its approach and run another experiment without breaking its flow.

Who can use GPT-5.6 Sol Ultrafast, and what is known about pricing

GPT-5.6 Sol on Ultrafast is a limited preview for a select group of customers, and OpenAI says it will expand access as capacity grows. Businesses that need frontier intelligence at the highest speed can use OpenAI's sign-up form to be notified when access expands.

On pricing, the announcement states no figure for GPT-5.6 Sol on Ultrafast. The API pricing page also lists no Ultrafast price for GPT-5.6 Sol as of October 10, 2026.

Ultrafast prices quoted on other sites do not come from OpenAI's announcement or pricing page, so none are used here. The prices published for the model itself are covered in GPT-5.6 Sol, Terra and Luna: Specs and Pricing.