drforbin.ai Submit a torrent

Public index · open-weight AI models · BitTorrent

Open weights.
Distributed freely.

The public index for AI model torrents. Magnet links and rich metadata for openly licensed language, image, audio, and embedding models — every listing verified against its official model card. We index. We never host.

12 listings · 7 categories · 7.35 TB indexed · 0 files hosted

magnet:?xt=urn:btih:6505b448ff71f38ff9b5a8025cf29cd6d2073b39&dn=deepseek-v4-1-flash
content address — verifies every byte infohash — the model's permanent ID display name — cosmetic only
The Week in Open Weights
Ternary Bonsai 2 puts a 27B in 5.9 GB, and the fine print matters

PrismML shipped Ternary Bonsai 2 27B on 17 September: Qwen3.8-27B with ternary weights, 5.9 GB on disk, Apache-2.0, and a headline claim of 98.2% retained performance. The claim is true as an average and misleading on the tasks the demo videos show, which is exactly the kind of number this community needs to read twice. Meanwhile the political week was dominated by fallout from Dario Amodei's "Pace the Frontier" essay, a Mozilla report putting open models 4.4 months behind the closed frontier, and Xi Jinping pitching a BRICS open-source AI community from New Delhi.

Read this week's edition →

Featured

Browse all →

DeepSeek-V4.1-Flash Quantized (FP8 weights / FP4 experts / FP4 KV cache)

552B (MoE, 8B prefill / 16B decode active) params · 510.31 GB · ctx 1024K

drforbin.aiMIT

GLM-5.3 BF16 safetensors (full precision)

753B (MoE) params · 1.51 TB · ctx 1024K

drforbin.aiGLM-5.3 License

Latest listings

RSS →

DeepSeek-V4.1-Flash Quantized (FP8 weights / FP4 experts / FP4 KV cache)

552B (MoE, 8B prefill / 16B decode active) params · 510.31 GB · ctx 1024K

drforbin.aiMIT

GLM-5.3 BF16 safetensors (full precision)

753B (MoE) params · 1.51 TB · ctx 1024K

drforbin.aiGLM-5.3 License

Kimi K3 MXFP4 release weights

2.8T (MoE, 104B active) params · 1.56 TB · ctx 1024K

drforbin.aiKimi K3 License

GLM-5.2 BF16 safetensors

753B (MoE) params · 1.51 TB · ctx 1024K

drforbin.aiMIT

Mixtral 8x22B Original release weights

141B (MoE 8×22B) params · 281.24 GB · ctx 64K

officialApache-2.0

Categories

Large Language Models6 listingsImage Generation0 listingsVideo Generation3 listingsAudio & Speech0 listingsMultimodal3 listingsEmbeddings0 listingsOther0 listings

Why torrents for AI models?

Weights are multi-gigabyte artifacts with bursty demand and long archival tails — exactly the shape BitTorrent was built for. No single point of failure, no bandwidth bill for creators, faster when it matters most.

Read the case Submit a torrent