Inference

Model serving that scales with your traffic

Deploy open and custom models behind a single OpenAI-compatible endpoint. Meridian handles GPU allocation, autoscaling, and health checks - you send requests.

See inference pricing Quickstart

~120ms p50 first-token latency

99.9% uptime target

0 clusters to manage

Capabilities

Everything between your model and your users

OpenAI-compatible API

Point your existing client at our base URL. Chat completions, embeddings, and streaming work out of the box.

Model catalog

Serve popular open-weight models - Llama, Qwen, Mistral families - or bring a fine-tuned checkpoint.

Autoscaling

Capacity scales with request volume, including scale-to-zero on idle endpoints so you stop paying when traffic stops.

Observability

Per-request metrics: latency, token throughput, error rates. Exportable to your monitoring stack.

Multi-region serving

Endpoints close to your users, with automatic failover between regions.

Usage-based billing

Pay per token or per request-minute. No idle GPU bills, no reserved-capacity contracts.

Quickstart

First request in under a minute

curl

$ curl https://api.meridian.finimble.com/v1/chat/completions \
  -H "Authorization: Bearer $MERIDIAN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "llama-3.1-70b",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Python

$ pip install openai

from openai import OpenAI

client = OpenAI(
    base_url="https://api.meridian.finimble.com/v1",
    api_key=os.environ["MERIDIAN_API_KEY"],
)
resp = client.chat.completions.create(
    model="llama-3.1-70b",
    messages=[{"role": "user", "content": "Hello"}],
)

Put your model behind a production endpoint

Start with usage-based pricing, or talk to us about dedicated capacity.

View pricing Contact sales