Pricing
Rates below are per 1M tokens and are the same for everyone. Signing up and minting an API key cost nothing; you need credit on the account before your first billable request, and the dashboard shows your balance plus any free allowance your organization has.
Call these at api.fabriqnetwork.com with your API key, metered per token and drawn from your credit balance. Only Live models are loaded on a node right now; a Loadable id answers 404, naming what is being served, until an operator loads it.
| Model | Family | Params | Context | Type | Status | Input / 1M | Output / 1M |
|---|---|---|---|---|---|---|---|
llama-3.2-1b | Llama | 1B | 128K | Text | - | $0.50 | $1.50 |
llama-3.2-3b | Llama | 3B | 128K | Text | - | $0.50 | $1.50 |
llama-3.1-8b | Llama | 8B | 128K | Text | - | $0.50 | $1.50 |
llama-3.1-70b | Llama | 70B | 128K | Text | - | $0.50 | $1.50 |
llama-3.3-70b | Llama | 70B | 128K | Text | - | $0.50 | $1.50 |
deepseek-v3.1 | DeepSeek | 671B MoE | 128K | Text | - | $0.50 | $1.50 |
deepseek-v3.2 | DeepSeek | 671B MoE | 128K | Text | - | $0.50 | $1.50 |
qwen3-235b | Qwen | 235B MoE | 256K | Text | - | $0.50 | $1.50 |
qwen3-vl-4b | Qwen | 4B | 128K | Vision | - | $0.50 | $1.50 |
kimi-k2-thinking | Kimi | 1T MoE | 256K | Reasoning | - | $0.50 | $1.50 |
glm-4.7 | GLM | 355B MoE | 128K | Text | - | $0.50 | $1.50 |
gpt-oss | OpenAI OSS | 20B | 128K | Text | - | $0.50 | $1.50 |
gpt-oss-20b | OpenAI OSS | 20B | 128K | Text | - | $0.50 | $1.50 |
gpt-oss-120b | OpenAI OSS | 120B | 128K | Text | - | $0.50 | $1.50 |
Prices are per 1M tokens. Models without a custom price use the network default ($0.50 in / $1.50 out).
Every request is metered on the tokens it actually used. No seats, no subscription, no idle capacity to rent.
Top up with USDT on Arbitrum One and spend it down. A platform fee applies to top-ups (currently 4%); the dashboard shows the exact credit you receive before you send anything.
Set a monthly spend cap in the dashboard and the gateway stops accepting requests once you reach it, instead of letting a runaway loop spend your balance.
Run a node and you accrue credit for the compute you serve. Payouts turn on as verifiable inference rolls out, so nobody is paid for unverifiable work yet.
Questions about a workload that doesn't fit this? support@fabriqnetwork.com