# Ai2 releases Olmo-core 3 with redesigned training for large AI models

_Published Thursday, October 1, 2026 at 11:07 AM EDT · AI, Infrastructure · Latest · Tier 2 — Notable_

![Ai2 releases Olmo-core 3 with redesigned training for large AI models — Primary](https://www.datocms-assets.com/64837/1790798606-olmo-core-3-blog-draft-google-docs-image-1.png)

Ai2 released Olmo-core 3, an open training framework for mixture-of-experts models, which activate selected specialized components for each input. The redesign keeps those components on GPUs and routes data to them, avoiding repeated gathering of model weights.

In a preliminary test on eight NVIDIA B300 GPUs, Ai2 said a 47-billion-parameter model processed 52,000 tokens per second per GPU, compared with 19,400 using its earlier implementation. Researchers can use and adapt the stack for their own models.

Ai2 also benchmarked a 1.2-trillion-parameter configuration across 512 GPUs. Those tests used random routing to measure system performance rather than trained-model quality; a separate 2.38-trillion-parameter configuration was a short capacity test.

## Sources

- [Ai2 Blog](https://allenai.org/blog/olmocore3)

---
Canonical: https://techandbusiness.org/newswire/C2dXPqPfWvTWZdE0ql1_9t
Published: 2026-10-01T15:07:02.146Z
Story chronology: 2026-10-01T08:00:00.000Z
Retrieved: 2026-10-01T18:17:57.357Z
Publisher: Tech & Business (techandbusiness.org)