Rent a dedicated instance of any open-weights model — Llama, Qwen, Gemma, DeepSeek and more — behind a single OpenAI-compatible endpoint. Billed per running minute, with keys you control.
# pip install openai from openai import OpenAI client = OpenAI( base_url="https://api.flexier.io/v1", api_key="sk-ac-live-…", ) stream = client.chat.completions.create( model="qwen3-8b", messages=[{"role": "user", "content": "Ship it."}], stream=True, )
~62 models — from a small model you can stream in real time to 8×-GPU giants — plus embeddings. Rent one as a private instance and address it by name.
3.1 / 3.3 · 1B → 405B
3 · 0.6B → 235B MoE
3 · 1B → 27B IT
V3 · R1 · distills
K2 · Moonshot · 8× GPU
Tencent · 7B → Large
Zhipu · 9B → 32B
M2.7 · 8× GPU
Qwen3 · Gemma · GTE
Everything you need to run production inference — and nothing you don’t.
Rent a dedicated GPU instance of any catalog model. Metered per running minute — provisioning is free, and a stopped instance costs nothing.
Put several of your same-model instances behind one name; the gateway balances requests across the least-loaded ready replica.
Point the OpenAI SDK at one /v1 base URL. Chat completions, streaming and embeddings — no code rewrite.
Prompt any running instance or group straight from the console — pick a target, stream a reply, no keys or curl.
Issue keys per organization with per-key rate limits, scope them to specific instances or groups, and revoke instantly.
A billing-grade runtime ledger with per-instance and daily breakdowns — see exactly what you spend, down to the minute.
Create your organization, mint a key, and make your first call in under two minutes.