Models and prices
Pay by the million tokens. See what each model needs from a host, how many hosts have it loaded right now, and how fast the first token arrives.
Call any of them
curl https://ballast.outerscope.xyz/v1/chat/completions \
-H "Authorization: Bearer $BALLAST_API_KEY" \
-d '{"model":"llama-3.1-8b-instruct",
"messages":[{"role":"user","content":"Hi"}]}'Hosts online
1
home GPUs with a fresh heartbeat
Tokens served today
172
finished requests since 00:00 UTC
Models listed
3
every one has a price above
Counts and first-token times are live. They refresh every 15 seconds.
| Model | Min VRAM | Input per M | Output per M | Hosts loaded now | First token, 1 h |
|---|---|---|---|---|---|
| Llama 3.1 8B Instructllama-3.1-8b-instruct | 8 GB | $0.05 | $0.10 | 1 | 421 ms |
| Qwen3 14Bqwen3-14b | 16 GB | $0.12 | $0.30 | No host online | — |
| Qwen3 32Bqwen3-32b | 24 GB | $0.25 | $0.55 | No host online | — |
Prices are dollars per million tokens. A model with no host online is still listed; a call to it returns no_capacity and nothing is billed.
Tokens in and out, at the prices above. A call that fails on one host is retried on another and not billed twice.
You buy credits first and spend them down. Payments run in Stripe test mode, so no real charge is made.
Point any OpenAI SDK at the base URL in the call above. GET /v1/models lists the same catalog as this page.