{"apiVersion":"v1","taxonomyVersion":1,"generatedAt":"2026-10-29T14:06:37.649Z","data":{"recordId":"wsa_9n864cys9yhc","score":1,"confidence":"high","capped":false,"capReason":null,"sources":[{"n":1,"url":"https://allenai.org/blog/olmocore3","publisher":"Ai2","authority":"owner-controlled","verification":"verified","lastRetrievedAt":"2026-10-28T00:27:38.878Z","detectedPublishedAt":"2026-10-01T08:00:00.000Z","rights":"original"},{"n":2,"url":"https://allenai.org/blog/colm-2026-recap","publisher":"Ai2","authority":"owner-controlled","verification":"verified","lastRetrievedAt":"2026-10-28T00:27:38.878Z","detectedPublishedAt":"2026-10-09T08:00:00.000Z","rights":"original"},{"n":3,"url":"https://huggingface.co/blog/allenai/olmocore3","publisher":"Hugging Face","authority":"other","verification":"verified","lastRetrievedAt":"2026-10-28T00:27:38.878Z","detectedPublishedAt":"2026-10-01T15:01:43.000Z","rights":"original"}],"claims":[{"id":"d67ec309-efd7-473e-b701-7ec14f4c3736","kind":"capability","statement":"Olmo-core 3 is designed to scale MoE training into the trillion-parameter range.","citations":[{"source":1,"supportProbability":0.99,"supports":true,"quote":{"text":"Olmo-core 3 is designed to scale MoE training into the trillion-parameter range while preserving computational efficiency.","rights":"original"}},{"source":3,"supportProbability":0.99,"supports":true,"quote":null}]},{"id":"eb3ca722-43e1-40b2-ae1b-3eb5bf189ebc","kind":"capability","statement":"A benchmark increased the expert pool from 8 to 128 while training throughput fell by less than 5%.","citations":[{"source":1,"supportProbability":0.98,"supports":true,"quote":{"text":"In one benchmark, we increased the expert pool from 8 to 128 while still selecting only four experts per token – the small units of text a language model processes – keeping the number of active parameters per token roughly fixed at about 3.2B.","rights":"original"}},{"source":3,"supportProbability":0.98,"supports":true,"quote":null}]},{"id":"eb24c68d-0b6a-42fd-9bc8-9eea0ac02790","kind":"capability","statement":"A preliminary test on eight NVIDIA B300 GPUs reported 52,000 tokens per second per GPU for a 47-billion-parameter MoE.","citations":[{"source":1,"supportProbability":0.98,"supports":true,"quote":{"text":"In a preliminary test on eight NVIDIA B300 GPUs, a 47-billion-parameter MoE processed 52,000 tokens per second per GPU with the new stack, compared with 19,400 using our earlier implementation—about 2.7× the throughput.","rights":"original"}},{"source":3,"supportProbability":0.98,"supports":true,"quote":null}]},{"id":"37d6d776-c204-4690-8148-235c64d85343","kind":"capability","statement":"A benchmark tested a 1.2-trillion-parameter model across 512 NVIDIA B300 GPUs.","citations":[{"source":1,"supportProbability":0.96,"supports":true,"quote":{"text":"We’ve benchmarked Olmo-core 3 across a range of configurations on NVIDIA B300 GPUs, including a 1.2-trillion-parameter model with 58.36 billion parameters active per token across 512 GPUs.","rights":"original"}},{"source":3,"supportProbability":0.95,"supports":true,"quote":null}]},{"id":"4a78fecd-8384-451a-933b-fca8a75dc12b","kind":"availability","statement":"Ai2 released Olmo-core 3 and linked to its code on GitHub.","citations":[{"source":1,"supportProbability":0.79,"supports":true,"quote":null},{"source":3,"supportProbability":0.65,"supports":false,"quote":null}]},{"id":"fe2ceb51-39a2-4452-aaff-7f3088ce619a","kind":"recommendation-relevant","statement":"Researchers and developers can use the open stack to train their own MoEs and adapt it to different hardware.","citations":[{"source":1,"supportProbability":0.98,"supports":true,"quote":{"text":"And it’s fully open—researchers and developers can use Olmo-core 3 to train their own MoEs, adapt it to different hardware, and experiment with routing, parallelism, and other parts of the system.","rights":"original"}},{"source":3,"supportProbability":0.98,"supports":true,"quote":null}]},{"id":"17427438-ada6-49c3-82a0-86fa588c76b8","kind":"availability","statement":"Olmo-core 3 was released ahead of COLM.","citations":[{"source":2,"supportProbability":0.98,"supports":true,"quote":{"text":"Released ahead of COLM, Olmo-core 3 adds a redesigned system for training large mixture-of-experts models.","rights":"original"}}]},{"id":"7a604f28-f950-4446-83fc-73bcd0d5621b","kind":"capability","statement":"Olmo-core 3 adds a redesigned system for training large mixture-of-experts models.","citations":[{"source":1,"supportProbability":0.97,"supports":true,"quote":null},{"source":2,"supportProbability":0.99,"supports":true,"quote":{"text":"Released ahead of COLM, Olmo-core 3 adds a redesigned system for training large mixture-of-experts models.","rights":"original"}},{"source":3,"supportProbability":0.97,"supports":true,"quote":null}]},{"id":"6150963b-6f64-4e49-82fa-f6d6b291b8ed","kind":"capability","statement":"Researchers can use the code to develop their own models.","citations":[{"source":1,"supportProbability":0.91,"supports":true,"quote":null},{"source":2,"supportProbability":0.98,"supports":true,"quote":{"text":"Researchers can use the code to develop their own models and investigate training decisions, alongside the checkpoints and technical reports available through the broader Olmo ecosystem.","rights":"original"}},{"source":3,"supportProbability":0.89,"supports":true,"quote":null}]}]},"nextCursor":null,"license":{"catalog":{"name":"CC BY 4.0","spdx":"CC-BY-4.0","url":"https://creativecommons.org/licenses/by/4.0/","attribution":"wowsuchai","attributionUrl":"https://wowsuchai.withglint.app","covers":["catalog metadata","classifications","generated summaries","feed event metadata"]},"excluded":{"covers":["source quotations","linked source content"],"terms":"Not licensed by wowsuchai. Source quotations and linked source content keep their original rights."}}}