openark

next save

gpt-oss 20b

0% of 11.14 sol

archive wallet
BnDENj…kmg9Dr

the decentralized hugging face.every trade saves a free ai model.

open source ai shouldn't live on one company's servers. $ARK fees fill a save bar per model, and a full bar puts it on arweave for good.

see the queuenext: gpt-oss 20b
  • paid for storage nov 13, 8:09 am-0.528 sol
  • drip in nov 13, 8:08 am+0.600 sol
  • archive wallet now0.072 sol
  • price to keep 1 gb forever$65.83
  • whole queue1.08 tb, $70.8k
  • saved so far1 model, 639 mb
  • times in miami time
saving to arweave8 models · 1.08 tb in line

now filling

gpt-oss 20b

openai, apache-2.0. 13.8 gb on hugging face. costs $907.76 to keep forever.

  • model-00001-of-00002.safetensors 4.8 gb sha256 4fbe328ab4
  • model-00000-of-00002.safetensors 4.8 gb sha256 16d0f997dc
  • model-00002-of-00002.safetensors 4.2 gb sha256 a18106b209
  • tokenizer.json 28 mb sha256 0614fe83ca
  • model.safetensors.index.json 36 kb git ae08521471
  • + 7 more files

after this

  1. qwen3 14b9.0 gb$592.59
  2. mistral 7b instruct14.5 gb$954.47
  3. qwen3.8 27b55.6 gb$3.66k
  4. glm-4.7 flash62.5 gb$4.11k

0%

0.087 of 11.14 sol in the archive wallet

11.05 sol to go

1 model, 639 mb saved forever.

how a save works

  1. fees come in.

    the dev drips sol from the coin's fees into the archive wallet. every drip is public.

  2. the bar fills.

    each model costs its size times arweave's live price. the front bar fills first.

  3. the files stream over.

    at 100% the files go straight from hugging face to arweave, no copy in between.

  4. every hash is checked.

    each file's fingerprint has to match the one hugging face publishes, or the save stops.

  5. it stays up for good.

    arweave is paid once and keeps the file. the links and hashes go on the record below.

how a save runs: qwen3 0.6b, 639 mb, $42.10walkthrough, not live

0.000

sol a full bar needs

sol in the archive wallet

  • qwen3-0.6b-q8_0.gguf639 mb
  • license12 kb
  • readme.md6 kb

hugging face 9465e63a22add5354d9bb4b99e901170
our stream  

the queue, smallest to biggest.

open models only, apache or mit, from the labs' own repos. bar height grows with size. the tallest is the big goal.

8 models
1.08 tb
$70.8k

saved forever.

read straight from arweave, not from us. each file links to its permanent copy.

check a file yourself.

pick a file, then hash a copy you downloaded or the copy on arweave. your browser does the math and compares it to hugging face's own fingerprint. no wallet, no upload.

pick a file to see its fingerprint.