Stanford Marin releases all 57 intermediate checkpoints of its 535B-A23B pretraining run
MaziyarPanahi · x · 2026-11-18
Stanford's Marin project has released all 57 intermediate checkpoints from pretraining its "Mochi" 535B-A23B model, including 10 from the long-context extension phase.
- Trained on 18.2T tokens
- Roughly 61 TB of weights, hosted on Hugging Face
- Accompanied by a detailed blog post
Open-sourcing the full checkpoint trajectory from a public research team is rare and valuable for scaling studies, loss-curve analysis, and ablation work — community members are calling it gold for open research.
More from Models
- GPT-6.2 Sol tops browser-use leaderboard at 82%, 35% cheaper and 2x faster than predecessor — charliermarsh · 2026-11-18
- Decision models beat LLMs at structured output: Liquid AI d1 runs 2.7-3.6x faster than LFM3 — mervenoyann · 2026-11-18
- Why Anthropic matters: competition, pricing and open source — dh7net · 2026-11-18
- GitHub to Retire Six Copilot Models on Dec 14, 2026, Auto-Switching Enterprise Users — rcleffke · 2026-11-18
- Dev's Deep Dive: GPT-6.2 Astra Improved but Still 'Directionally Very Bad' at Debugging — kimmonismus · 2026-11-18
- OpenAI Charged for a ChatGPT Business Premium Seat for a Month That Can't Be Used — physmathastro · 2026-11-18