The Deploy Log | Breakthroughs
Olmo-core 3
Ai2 released Olmo-core 3, an open MoE training system benchmarked at over one trillion total parameters, with a 47-billion-parameter MoE processing 52,000 tokens per second per GPU on eight NVIDIA B300 GPUs, about 2.7x the throughput of the earlier implementation.
Allen Institute for AI | In Edition 54 , Saturday 3 Oct 2026
SOURCE Allen Institute for AI, Introducing Olmo-core 3: Open, scalable training infrastructure for large MoEs
Added to The Deploy Log Sunday 4 Oct 2026 . Read Edition 54 , the edition that carried it.
Week of September 28, 2026 October 2026
One line every time AI actually ships, as it ships, 24/7. Follow the wire: Threads | Bluesky
The call on this one
WATCH if you train large MoEs
More from Allen Institute for AI
Everything Allen Institute for AI has shipped | All The Deploy Log
One line every time AI actually ships, around the clock, free on Threads and Bluesky. Every one is in The Deploy Log with its source.
Follow on Threads Follow on Bluesky
The Saturday edition is the week in one, free to read on the site, and by email if you want it: get it by email .