Checked fact 172317 Oct 2026Models and releases

The model card says the design goal is to let users trade inference compute for accuracy by changing the loop count, from one loop up to four.

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

By changing how many times the shared block of layers is applied, you can trade inference compute for accuracy to suit each task, with no retraining and no model swap.

What the source said when we opened it, on 17 Oct 2026.

The source

NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16 model card
Hugging Face / NVIDIA

Open the source archived copy

Checked

Checked by the notis newsroom on 17 Oct 2026, against the source above.

In the story

Nvidia listed Dory 1.0 4B, a looped model with 256K context and up to 61.8 on SWE-bench Verified. 17 Oct 2026

Cite this fact

Short permanent address

In a sentence

“The model card says the design goal is to let users trade inference compute for accuracy by changing the loop count, from one loop up to four.” ai notis, checked 17 Oct 2026. Source: NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16 model card (Hugging Face / NVIDIA). https://ainotis.com/c/1723

As a reference (APA style)

ai notis. (2026, October 17). The model card says the design goal is to let users trade inference compute for accuracy by changing the loop count, from one loop up to four [Checked claim 1723]. ai notis. https://ainotis.com/c/1723

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