Pricing

AICF compute pricing

Prices on this page are read from the live AICF model registry, not typed in by hand. They are quoted in ANM per unit; the registry stores them as integer nANM. The final rate for a key is shown at key creation on console.animica.org and governs if the two ever differ.

Get an API key Use the free tier Provider economics

Three ways to pay (or not)

Tier	Auth	Limits	Notes
Free (animica.dev/v1)	No key, no signup	30 requests per minute per IP	Funded by the foundation treasury; served by the miner network; ML-DSA-65 proof-of-inference receipts
Paid key (console.animica.org)	API key, project budget	Per-plan rate limits	Billed from an ANM balance at the registry rates below
Per-call (x402.animica.dev)	None; pay per request	Per product	HTTP 402 flow; ANM lane (animica:1) or USDC lane; price is in the 402 response

Most people should start with the free endpoint. It is OpenAI-compatible, so existing client libraries work by changing the base URL to https://animica.dev/v1 and omitting the key. Capacity is community GPU: check each model's serving flag in /v1/models, because a model with no worker online can queue or return 503. Paid keys exist for budgets, higher rate limits and per-project accounting; x402 exists for autonomous agents that cannot hold an account.

Registry rates

The table is generated from GET https://aicf.animica.org/api/models and was last re-verified against the live registry on 2026-08-23. Rates are indicative; the registry is the source of truth.

Item	Unit	Base price	Provider share	Treasury share
Chat Input (aicf-chat-1)	1K input tokens	`0.012 ANM`	85%	15%
Chat Output (aicf-chat-1)	1K output tokens	`0.018 ANM`	85%	15%
Chat Request Base Fee	request	`0.00015 ANM`	85%	15%
Embeddings Input (aicf-embed-1)	1K input tokens	`0.008 ANM`	82%	18%
Embedding Vectors (aicf-embed-1)	vector	`0.00005 ANM`	82%	18%
Embeddings Request Base Fee	request	`0.0001 ANM`	82%	18%

Read the registry yourself

    curl -s https://aicf.animica.org/api/models | jq '.data[] | {id, pricing}'

# Excerpt (2026-08-23): values are nANM per unit
# "inputTokenAnmNanos":  "12000"    ->  0.000012 ANM / input token  = 0.012 ANM per 1K
# "outputTokenAnmNanos": "18000"    ->  0.000018 ANM / output token = 0.018 ANM per 1K
# "requestBaseAnmNanos": "150000"   ->  0.00015  ANM / request
# "providerShareBps": 8500, "treasuryShareBps": 1500

A worked example

A chat completion on aicf-chat-1 with a 1,000-token prompt and a 300-token answer:

aicf-chat-1, 1,000 in / 300 out

    1,000 input tokens × 0.000012 ANM  = 0.012  ANM
  300 output tokens × 0.000018 ANM = 0.0054 ANM
  request base fee                 = 0.00015 ANM
  ------------------------------------------------
  total                            = 0.01755 ANM  (17,550,000 nANM)
  provider 85%                     = 0.0149175 ANM
  treasury 15%                     = 0.0026325 ANM

At the ANM/USDT rate on NonKYC the request costs a small fraction of a cent in USD terms; we do not quote a dollar figure here because it changes continuously. What does not change without a registry update is the ANM rate and the split.

How the model works

Pricing is per model, not per provider. Each registry entry carries an input-token rate, an output-token rate (or a per-vector rate for embeddings), a per-request base fee, and the provider/treasury split in basis points. When a job settles, the provider share is credited to the worker that served it and the treasury share to the AICF treasury. The split is what funds the free tier and trap-prompt evaluation; it is also why providers see 82 to 85% of list price rather than all of it. Separately from per-job billing, 25% of each block subsidy has been earmarked for inference providers since block 75,000; that subsidy is a reward for serving, not a charge to users, and is described on the Providers page.

The node software also ships a general billing module (billing/, described in docs/monetization.md) with a free mode (the default: no key, rate-limited, zero fees) and a paid mode with an x-animica-key header, per-plan rate limits (free 60, pro 600, enterprise 6,000 requests per minute by default), optional per-byte DA fees and flat RPC fees. The public RPC at rpc.animica.org runs in free mode; those knobs exist for operators hosting their own paid infrastructure.

What this page does not tell you

Prices for x402 products are not listed here; each product's 402 response is authoritative and can change with gas conditions on the USDC lane.
There is no published SLA for the free tier. Paid keys are governed by the terms shown in the console.
Training, fine-tuning and batch job classes are described in the AICF docs as asynchronous job types; they have no fixed public rate on the registry today, so none is shown.

Sources

live aicf.animica.org/api/models (2026-08-23) · website/src/data/aicf.ts · docs/monetization.md · docs/aicf/SLA.md · docs/ANIMICA_2026_STATE.md · docs/x402.md