GPT-6 now available

GPT-6 is now live in Saga, replacing GPT-5.5 and GPT-5.6 Sol/Terra/Luna.

Long-context reliability has improved dramatically, which is relevant for working through data rooms and large contract sets. GPT-6 has an increased context window, and on OpenAI's needle-in-a-haystack test at 512K to 1M tokens it scores 96.3%, up from 73.8% for its predecessor. That is the difference between a model that technically accepts a full due diligence file and one that can actually find what's in it.

OpenAI reports that GPT-6's rate on its internal hallucination evaluation fell to 4.2%, down from 12.2% for GPT-5.6 Sol. It is also three times less likely to misrepresent its own capabilities, and in OpenAI's boundary tests it never attempted to get around an automated review denial, even when the task could not be completed any other way. For anyone deploying AI in a regulated line of work, a model that knows its limits and stays within them is exactly what you want.

On data handling, GPT-6 supports Zero Data Retention for eligible API customers, which is what allows us to run it in Saga under the same confidentiality standards as every other model on the platform.

In the ongoing battle between OpenAI and Anthropic, OpenAI has struck back quickly, and the gap at the frontier is as narrow as it has ever been. For our users, that competition simply means better tools.

GPT-6 is available now in Saga.

‍