SLQS001-JOBSiliq datasheet

Buy answers and frames from real GPUs.

Post a prompt and a miner's GPU answers it, or bring your own key for Claude and DeepSeek. Pay ETH for a key without the hourly limit or for rendered frames. That ETH refills the vault that pays miners.

Prompts answered by miners' GPUs.

Post a prompt and pick an open model. A miner's tab running SLQ-05 with those weights loaded, or a headless node, answers on its own GPU, and the response names the card. Decoding is greedy, so the same prompt gives the same tokens on a second GPU. Free for 24 prompts an hour per address, up to 256 tokens each.

GPUs serving now
Answers, 24 h
Tokens, 24 h
Answers, all time
Rechecks matched
Rechecks mismatched

Open models, where they run, credits per prompt and GPUs serving each one now
Weights	Params	Runs in	Size	Credits per prompt
SmolLM2-135M-Instruct	135M	miner tab, SLQ-05	0.12 GB	1
SmolLM2-360M-Instruct	362M	miner tab, SLQ-05	0.27 GB	2
SmolLM2-1.7B-Instruct	1.7B	miner tab, SLQ-05	1.1 GB	8
DeepSeek-R1-Distill-Qwen-1.5B	1.8B	miner tab, SLQ-05	1.3 GB	8
Qwen3-8B	8.0B	headless node	5.2 GB	6

rate Credits apply to paid keys only. The free tier costs nothing. Serving counts refresh every 10 seconds.

POST /api/v1/inference

# SILQ is the origin of this site
export SILQ=https://<this site>

curl -X POST $SILQ/api/v1/inference \
  -H 'content-type: application/json' \
  -d '{"prompt": "What does a foundry do?",
       "model": "smol-360m", "wait": false}'

# 202 with the job id, then poll it
curl $SILQ/api/v1/inference/<id>
{ "id": "<id>", "status": "done", "model": "smol-360m",
  "output": "...", "tokens": <int>, "ms": <int>,
  "gpu": "<gpu label>", "served": "miner", "verified": null }

"wait": falseReturns 202 with the job id at once. Without it the call holds up to 25 s, then returns 202 with a poll URL if no GPU has finished.
GET /api/v1/inference/{id}Reports queued, running with the partial text, or done with the GPU, tokens and milliseconds.
POST /api/v1/inference/{id}/verifyReplays a miner's answer on a second GPU, 10 times an hour per address. The verdict lands in verified.
$SILQThe origin of this site. The first line of each block sets it. Any OpenAI client works with base_url $SILQ/api/v1, and CORS is open.

Ask a miner

POST /api/v1/inference

The answer lands here, with the GPU that wrote it.

Frontier models on your own key.

Claude and the large DeepSeek models do not fit in a tab or on a home node. Siliq passes your prompt to the maker's API with the key you send in the x-provider-key header. The key serves that one call and is never stored. The provider bills you at its list price, and Siliq adds no markup.

Prompt cap rate: 20,000 chars
Per address rate: 60 an hour

Try them in the chat

Frontier models that run on the caller's own provider key
Model id	Name	Weights	Billing
claude-opus-5	Claude Opus 5	closed	Anthropic list price, billed to your key
claude-sonnet-5	Claude Sonnet 5	closed	Anthropic list price, billed to your key
deepseek-chat	DeepSeek Chat	open, too large for a tab	DeepSeek list price, billed to your key
deepseek-reasoner	DeepSeek Reasoner	open, too large for a tab	DeepSeek list price, billed to your key

POST /api/v1/inference with x-provider-key

# SILQ is the origin of this site
export SILQ=https://<this site>

curl -X POST $SILQ/api/v1/inference \
  -H 'content-type: application/json' \
  -H 'x-provider-key: sk-...' \
  -d '{"prompt": "What does a foundry do?",
       "model": "claude-sonnet-5"}'

# answered in the same request
{ "status": "done", "model": "claude-sonnet-5", "output": "...",
  "tokens": <int>, "ms": <int>,
  "served": "anthropic", "servedBy": "<provider model>" }

servedBy is the model the provider says wrote the answer. DeepSeek Reasoner also returns its reasoning. No Siliq key is needed for these models.

Paid keys lift the hourly limit.

$1 of ETH buys 1,000 credits. A key has no hourly limit and answers up to 512 tokens per prompt. A prompt costs 1 to 8 credits by model, as the table above lists. Send at least 0.001 ETH on Robinhood Chain, sign with the wallet that sent it, and the key shows once.

Pay to
ETH pricemarket
Credits per ETHrate
Credits per $1rate: 1,000
Minimumrate: 0.001 ETH

Pay for frames. Miners render them.

Miners on SLQ-03 trace your scene while their run scores. The instrument is a path tracer, 4 bounces, seeded sphere field over a ground plane, and your seed decides the scene. The ETH you pay refills the vault.

Minimum rate: 0.002 ETH
Frames per ETH rate: 5,000
Per GPU rate: 1 frame every 15 s
Board keeps rate: last 24 frames

Order frames

Seed

The seed places the spheres. Same seed, same scene.

Title

Shown on the public board. Up to 48 characters.

Job board

refreshes every 15 s

Run a node.

A browser tab holds the small models. A headless node serves bigger ones from your own Ollama, pulls from the same queue, and labels each answer with its GPU. It registers by signing one message with a wallet key that stays on your machine and sends no transaction.

Nodes are not paid yet. A node answers prompts and earns nothing for now. Mining runs are what the vault pays. After launch, a node wallet must hold $25 of $SILQ.

Terminal, needs Node and Ollama

# SILQ is the origin of this site
export SILQ=https://<this site>

curl -O $SILQ/silq-node.mjs
curl -O $SILQ/silq-node.sha256 && sha256sum -c silq-node.sha256
# macOS: shasum -a 256 -c silq-node.sha256
npm i viem
ollama pull qwen3:8b
PRIVATE_KEY_FILE=~/.siliq/key node silq-node.mjs --models=qwen3-8b --base=$SILQ

PRIVATE_KEY_FILEA file holding one 0x private key. Keep the key in a file: a key typed inline lands in your shell history.
--modelsComma-separated Siliq model ids. A node can serve the tab models too.
--gpuThe label shown next to each answer, for example the card name.

Models that need a node
Model id	Ollama tag	Params	Size	Credits per prompt
qwen3-8b	qwen3:8b	8.0B	5.2 GB	6