Open-source lab Marin kicks off 535B/A23B MoE hero run, going 'boringly well'
BrandoHablando · x · 2026-09-04
Marin, the open-source training project under Open Athena, has started a 535B-parameter (23B active) MoE hero run that has been going almost boringly well for a few weeks, according to the author. Marking one year since Marin joined Open Athena, the author also published a write-up on how the fully open frontier-scale training effort got this far. With code, data and process all public, it is one of the few open-source attempts at frontier-scale pretraining and a notable reference point for the research community.
Related event: Stanford's Marin Launches Fully Open 535B-A23B MoE Training Run(5 posts)→
More from Models
- OpenAI launches controversial Astra model, starting with Daybreak cybersecurity customers — recoveringasshole0 · 2026-09-04
- GPT-6 Astra lands: 98.6% on ARC-AGI 3, Brockman declares the AGI era — Expert_Annual_19 · 2026-09-04
- OpenAI rolls out GPT-6 Astra to vetted cyber customers at 2.5x GPT-5.6 pricing — rohanpaul_ai · 2026-09-04
- Analyst doubles down: OpenAI set to out-accelerate Anthropic by year-end — scaling01 · 2026-09-04
- GPT-6 Astra reportedly scores 61.2 on AAII, slightly above GPT-5.6 Sol — Angaisb_ · 2026-09-04
- 12B local models now rival GPT-4, and tiered local AI is coming, argues Wardell — draginol · 2026-09-04