Summary
According to a post on X by @ClementDelangue, the Marin project released all 57 intermediate checkpoints from pretraining its "Mochi" 535B-A23B model, including 10 from long-context extension. Separately, RadixArk released its first open-weight model, DaisyRose-27B, which the post says was trained in 18 days on 144 H200 GPUs across 1,200 long-horizon agent tasks. The post also says Hugging Face is handing model inference traffic to OpenRouter. The source snippet is cut off, so the full context is not available here.
AI-assisted summary based on the listed source.
What happened
The Marin project released all 57 intermediate checkpoints from pretraining its "Mochi" 535B-A23B model, including 10 from long-context extension. RadixArk released its first open-weight model, DaisyRose-27B, trained in 18 days on 144 H200 GPUs across 1,200 long-horizon agent tasks, and Hugging Face is handing...
Why it matters
Access to intermediate checkpoints lets researchers and builders see how a very large model changes during training, not just its final weights. For teams building on open models, DaisyRose-27B is a new option to evaluate, especially for agent-style tasks. Developers who use Hugging Face inference may want to watch how the shift to OpenRouter is carried out.
Signal Intelligence
Signal Strength 90%
Technical label WATCH
Public Interest 51
Category OPEN SOURCE
Reader Depth TECHNICAL
Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.
Public Interest components
Recognizable Entity Score 51
Practical Impact Score 36
Novelty Interest Score 94
Consequence Score 30
Curiosity Score 16
Shareability Score 62
Why this is here
VQV surfaced this signal because it is recent, relevant to AI Chips, connected to Scobleizer News Pulse (X).