Fetching from the wire…
01
02
03
04
05
06
07
08
OSS2026-09-21 · source-backed
pirateface.co launched a torrent index mirroring open weights as magnet links, restricted to Apache-2.0 and MIT weights plus one approved exception for Kimi-K3. Every file carries its official Hugging Face SHA-256 fingerprint, so a copy from any peer has to match byte for byte, and the torrent keeps resolving after Hugging Face stops serving the file. 537 points on Hacker News September 20. Creator-claimed handles are verified against the matching HF account to block impersonation. The durability argument is the one I'd take seriously: open weights you can only get from one host aren't as open as they look.
Each link below shares sources, entities, or timing with this story.
The file is 1.56TB. That's the first thing you notice about the moonshotai/Kimi-K3 Hugging Face repo that went live today. 2.8 trillion total parameters, 104B activated, 896 routed experts with 16 selected plus 2 shared per token, 93 layers split 69 Kimi Delta Attention and 24...
$12,930,300,000. Not "about $13 billion." That exact figure is in Jensen Huang's September 3 post, and the precision is the first sign this is a signed agreement rather than the leak-stage reporting we got in late August. NVIDIA's blog frames the platform in the numbers that m...
TechCrunch strings together three deals: Nvidia's reported $13 billion Hugging Face acquisition, its $6 billion Poolside arrangement, and Stripe's acquisition of OpenRouter for over $7 billion about two weeks before August 28. The thesis is acquirers hedging against frontier-l...
This is a supply-chain fact, and most people are still treating it as a geopolitics argument. Sequoia published "America's Open-Model Paradox" on July 24 with the number that reframes the whole conversation: Qwen's share of open-model fine-tunes went from 1% in January 2024 to...
Salesforce released Koa, built by post-training Nemotron-3-Super-120B, Nvidia's open-weight hybrid Mamba-Transformer MoE with 120B total and 12B active parameters (TechCrunch, paper at arXiv 2609.15066). The training was GRPO reinforcement learning on public and synthetic data...
For about a year, "run your agent locally" meant accepting a model that couldn't reliably call a tool twice in a row. That excuse is gone. Meta Superintelligence Labs published Muse Glimmer today: a 29.6B dense causal transformer, 52 layers, 6,656 hidden dim, with a ~1.8B ViT-...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.