root layer
For developers · demand

batch-first,
drop-in inference.

Async, no latency SLA — so consumer-GPU variance stops mattering and the price drops. OpenAI-compatible: change one line, pay in fiat or crypto, and settle on compute you can verify.

Batch-first buyers

built for high-volume, not low-latency.

We picked the demand that consumer GPUs are perfect for: cheap, parallel throughput.

Async by design

batch-first api

No latency SLA — submit a job, collect results. Consumer-GPU variance stops mattering, so the price falls.

The workloads

high-volume inference

Embeddings, classification, JSON extraction, OCR, transcription — the jobs you run by the million.

Migration

drop-in, openai-compatible

Change one line — the base URL. Keep your SDK, your prompts, your pipeline. Pay in fiat or crypto.

Launch pricing

token-subsidized

Vested $RL cashback lowers your effective cost during the network bootstrap.

Mission demand

ai-for-good jobs

Grant-backed research and public-good workloads for early traction — and a way to give compute back.

Verifiable

a receipt for every job

Each result ships with cryptographic proof of exactly what model ran, on which node.

Change one line

point your client at root layer.

If you already call an OpenAI-compatible endpoint, you're one base-URL away. Same request shape, verifiable settlement underneath.

  • OpenAI-compatible REST & SDK surface
  • Idempotent batch jobs with webhooks
  • Pay-per-token, settled in stable credits
  • Every response carries a compute receipt
quickstart.py · python
from openai import OpenAI

client = OpenAI(
    base_url="https://api.rl.foundation/v1",
    api_key="rl-...",
)

resp = client.embeddings.create(
    model="bge-large-en",
    input=["verifiable compute, on idle GPUs"],
)

print(resp.data[0].embedding[:4])
print(resp.receipt.proof)   # proof of useful work
batch.sh · bash
curl https://api.rl.foundation/v1/batches \
  -H "Authorization: Bearer rl-..." \
  -d '{ "model": "whisper-small",
        "input_file": "calls.jsonl",
        "settle": "verified" }'
Model catalogue

smaller models, consumer-ready.

Smaller models now match older, larger ones — so a tiered catalogue runs cleanly on the cards the network is built from. Illustrative launch pricing.

ModelTaskContextPrice
bge-large-enEmbeddings512$0.006 / 1M tok
MiniLMClassification256$0.004 / 1M tok
whisper-smallSpeech-to-text—$0.05 / hr audio
FlorenceOCR / vision—$0.40 / 1k images
Llama-3.1-8BText generation8k$0.09 / 1M tok

Pricing illustrative — final rates set at mainnet against live network supply.

Verified compute

trust-minimized, not trust-us.

Migrate in1line

Point your existing OpenAI client at a new base URL.

Jobs re-checked100%

Staked validators duplicate and canary-check every batch.

Pay your wayFiator crypto

Settle in stable credits — no wallet gymnastics required.