Private endpoints your SDK already speaks
Swap your OpenAI or Anthropic base URL. The rest of your code stays the same. Run it on serverless, dedicated or reserved capacity.
from openai import OpenAI
client = OpenAI(
base_url="https://api.nucleus-ai.io/v1", # the one line that changes
api_key=NUCLEUS_API_KEY,
)
client.chat.completions.create(
model="nucleus://run_3Bz:train:0/sampler_weights/000240",
messages=[{"role": "user", "content": "Triage this ticket"}],
)Call your model the way you already call one
Change one line
Point your OpenAI or Anthropic client at Nucleus and keep the rest of your code.
Base model or your checkpoint
The model parameter takes a base model ID or the nucleus:// path of your fine-tuned weights, so one client serves both.
Pay for what runs
Serverless meters tokens and idle time costs nothing. A dedicated GPU bills by the second.
Capacity you can hold
Reserve a GPU for one to six months and the capacity is guaranteed.
Five ways in, one key
Point your OpenAI or Anthropic client at Nucleus and authenticate with your Nucleus key. The model parameter takes a base model ID or the nucleus:// path of your fine-tuned checkpoint. Both routes need a key with the infer scope.
Results from LLMs powered by Nucleus
daily fuel saved across the fleet
delays resolved without a call
from site reading to client report
average time to resolution
Models behind a live feed
47 trucks rerouted while they were still moving
Cable survey readings processed while the crew was still on site
Every shipment two days past ETA, found before the customer called
What the Serve column includes
Routes
- OpenAI-compatible chat completions and completions at /v1
- Anthropic-compatible messages and count_tokens at /anthropic
- Token-level sampling on /v1/samplers
- Streaming over server-sent events on completions, chat completions and messages
Models
- A base model ID or a nucleus:// checkpoint path
- Language, vision, embedding and reranking, image, video and audio groups
- A fine-tuned model served at its base model's rate
Capacity
- Serverless per token, idle costs nothing
- Dedicated L40S, A100, H100, H200 and B200 by the second
- Reserved one to six months, 15 to 25% below on-demand
- Cold start free
Keys and metering
- Keys carry scopes, and serving needs infer
- Authorization Bearer everywhere, x-api-key accepted on the Anthropic route
- The usage API reports inference tokens by day, project or base model
- Count tokens before you send
Two meters run the serving bill
Served tokens
Input and output tokens cost the same. Dedicated GPUs from $1.99 per GPU-hour, billed by the second. Reserved from $1.49.
Pick your path
Private Endpoints
OpenAI and Anthropic compatible
Serverless
Pay per token, idle costs nothing
Dedicated GPUs
L40S to B200, by the second
Reserved capacity
Guaranteed terms, one to six months
Compatible APIs
DocsOne key, two SDK families
Read on
Serving and exporting
Take finished weights to production by serving them on Nucleus or exporting them to run anywhere.
Use the Anthropic SDK and Claude Code
Point the Anthropic SDK or Claude Code at a Nucleus-served model with a base URL change.
Mandai Wildlife Group transforms animal care with Nucleus
How Mandai uses Nucleus AI to streamline animal care operations across their wildlife parks.
Change one line and call your model
Your OpenAI or Anthropic client works as it is. Pick the capacity that fits the traffic.