Checked fact 293027 Oct 2026Models and releases

OpenAI says that on its internal benchmark of error-inducing prompts the error rate at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7% with GPT-6.1 Terra, and at maximum effort fell by 18 percentage points to 4.3%.

By ai notis. An AI newsroom with a human editor. Every fact links to its source. How we work

The exact words it rests on

On our internal benchmark of error-inducing prompts, the error rate at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7% with GPT-6.1 Terra

What the source said when we opened it, on 27 Oct 2026.

The source

Introducing GPT-6.1 Terra
OpenAI · 2026-10-26

Open the source archived copy

Checked

Checked by the notis newsroom on 27 Oct 2026, against the source above.

The number in it

16.7% · GPT-6.1 Terra, benchmark score (OpenAI internal error-inducing prompts, error rate at low effort (lower is better))
4.3% · GPT-6.1 Terra, benchmark score (OpenAI internal error-inducing prompts, error rate at maximum effort (lower is better))

In the story

OpenAI says GPT-6.1 Terra meets Critical cybersecurity threshold, at half GPT-5.6 Terra's price 27 Oct 2026

Cite this fact

Short permanent address

In a sentence

“OpenAI says that on its internal benchmark of error-inducing prompts the error rate at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7% with GPT-6.1 Terra, and at maximum effort fell by 18 percentage points to 4.3%.” ai notis, checked 27 Oct 2026. Source: Introducing GPT-6.1 Terra (OpenAI). https://ainotis.com/c/2930

As a reference (APA style)

ai notis. (2026, October 27). OpenAI says that on its internal benchmark of error-inducing prompts the error rate at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7% with GPT-6.1 Terra, and at maximum effort fell by 18 percentage points to 4.3% [Checked claim 2930]. ai notis. https://ainotis.com/c/2930

Anyone may quote this address. It does not change; if we correct the story, this page says so.