ᚠ

Start free.
Pay only for what you use.

Try with 500 free requests a day. Deposit USDC on Base to unlock all 532 models — you can never overspend.

ᚱᚢτᚾᛊ
ᚨ
Free
$0
Try before you buy
  • 500 requests/day
  • Small models only (up to 70B)
  • Llama 3.3 70B, Mistral 7B, Qwen 7B, Gemma 9B, Phi-4
  • max_tokens capped at 1,000
  • No streaming
  • OpenAI-compatible format
  • No credit card or wallet required
Get Free Key →
ᛞ
Paid — Pay As You Go
USDC on Base
Deposit credits, pay per request
  • Unlimited requests
  • ALL 532 models — DeepSeek R1 671B, GPT-4o, Claude, 405B+
  • Streaming enabled
  • No token cap
  • 50% markup on upstream cost
  • Fine-tuning access ($15–75/hr)
  • Can never overspend — blocked at $0 balance
Deposit USDC →

Example per-request costs (with 50% markup)

Llama 3.3 70B
~$0.000075
per request
DeepSeek R1 671B
~$0.00045
per request
GPT-4o
~$0.003
per request
Fine-Tuning
$15–75
per hour
ᛟᚨτᛃᚾ

Frequently asked

How does billing work?
Deposit USDC on Base. Every API call deducts from your balance based on the model used. There is a 50% markup on upstream cost. No subscriptions, no monthly bills. You can never overspend — when your balance hits $0, paid models are blocked and you fall back to the free tier.
What if my credits run out?
Paid models return a 402 Payment Required response. You still have access to the free tier (500 req/day, small models). Top up anytime — credits appear in ~30 seconds.
Can I use it for free?
Yes. 500 requests/day with small models (up to 70B parameters) like Llama 3.3 70B, Mistral 7B, Qwen 7B, Gemma 9B, and Phi-4. No deposit needed, no streaming, max_tokens capped at 1,000. Just sign up and get an API key.
What models require a deposit?
All large models (405B+), proprietary models (GPT-4o, Claude), and frontier models (DeepSeek R1 671B) require credits. Streaming and uncapped token output also require the paid tier. Small open-source models up to 70B are available on the free tier.
How much does fine-tuning cost?
Fine-tuning costs $15–75/hr depending on model size (50% markup on upstream Gradients rates). Powered by Gradients (SN56) on Bittensor's decentralized GPU network. Check /v1/fine-tuning/prices for current rates. Requires the paid tier.
Can I overspend?
No. Credits are checked before every request. If your balance is below the estimated cost, the request is blocked with a 402 response. You can only spend what you have deposited. Top up whenever you want.