Five ways in

No account. No sign-up. Pick the method that works from where you are. Every path serves the exact same bytes, verified by the same signed manifest.

HTTPS

The fastest option for most people. Direct download from our edge servers spread across four countries.

Speed Fast — up to your line speed

Best for One-off downloads, small to medium models

Domains .org · .is · .ch · .nl

BitTorrent

Community-seeded. The more people seeding, the faster it gets. Each model has a magnet link on its download page.

Speed Depends on seeders — often very fast

Best for Large models, resuming interrupted downloads

You help Leave your client open to seed for others

IPFS

Content-addressed and permanent. The file's address is a hash of its content — you cannot seize a hash.

Speed Varies — faster with a local IPFS node

Best for Permanent archival, censorship resistance

URI ipfs://bafybei…

Tor .onion

For when you need the strongest privacy. Access the full catalog through Tor Browser — no clearnet connection at all.

Speed Slower — the cost of anonymity

Best for High-risk regions, maximum privacy

Address ommxyz…onion

CLI — `omm`

One command to search, download, and verify. Built for people who live in the terminal.

Speed Auto-selects the fastest available method

Best for Scripting, batch downloads, automated verification

Install curl -fsSL openmodelmirror.org/install.sh | sh

If nothing works

If all five methods are blocked from where you are, there are still ways in. The catalog is also pinned to IPFS, archived on Nostr, and distributed through community mirrors. Check our status page for live endpoint health.

As a last resort, the entire static site can be served from a USB stick or any local file server — every download link, every hash, every verification step still works with JavaScript off.

Example: download Llama 3.1 70B

# With the CLI — one command
omm download meta-llama-3.1-70b-instruct --verify

# With BitTorrent
magnet:?xt=urn:btmh:1220b94d27b9934d3e08a52e…

# With IPFS
ipfs get bafybeih5k7r…

# With HTTPS (pick any domain)
wget https://openmodelmirror.org/models/meta-llama-3.1-70b-instruct/

All four commands deliver the same 141.5 GB of verified model weights. Then verify — see the verification page.