One key. Every model.
The cheapest healthy provider.
UberLLM is an OpenAI-compatible inference marketplace. Swap one base URL and every request routes to the cheapest provider that is actually up — with automatic failover across the whole order book.
Prepaid credits · OpenAI & Anthropic SDKs · BYOK · Streaming
from openai import OpenAI
client = OpenAI(
base_url="https://api.uberllm.dev/v1", # ← the only change
api_key="ull_your_key_here",
)
resp = client.chat.completions.create(
model="deepseek/deepseek-v3.1",
messages=[{"role": "user", "content": "Explain routing in one line."}],
)
print(resp.choices[0].message.content)Reference prices, live
Pass-through provider pricing — the cheapest healthy endpoint wins.
| Model | Context | Input /MTok | Output /MTok | Providers |
|---|---|---|---|---|
Claude Haiku 4.5 anthropic/claude-haiku-4.5 | 200K | — | — | |
Claude Opus 4.8 anthropic/claude-opus-4.8 | 1000K | — | — | |
Claude Sonnet 5 anthropic/claude-sonnet-5 | 1000K | — | — | |
DeepSeek R1 deepseek/deepseek-r1 | 164K | — | — | |
DeepSeek V3.1 Terminus deepseek/deepseek-v3.1-terminus | 164K | — | — | |
DeepSeek V3.2 deepseek/deepseek-v3.2 | 164K | — | — |
One runtime, two sides
Stop shopping providers
- Change one base URL. Your existing OpenAI or Anthropic SDK just works.
- Cheapest healthy route. Inverse-square-of-price load balancing, 30s outage windows.
- Automatic failover. A dead provider is skipped before the first token streams.
- Prepaid credits. Per-key spend caps, itemized cost per request and model.
Monetize idle capacity
- List any OpenAI-compatible endpoint. Set an absolute price or a multiple of the reference.
- Get real traffic. You win requests whenever you are the cheapest healthy offer.
- Passive health only. We never burn your quota with uptime pings.
- Get paid per request. Settlements accrue to a payable balance you can withdraw.
Drop-in compatible
OpenAI /v1/chat/completions and Anthropic /v1/messages — same shapes, streaming included.
Smart routing
order · only · ignore · sort · max_price. Shortcuts :floor and :nitro.
BYOK
Bring your own provider keys — routed under your account, zero markup.
Transparent metering
Provider-reported usage, tokenizer fallback, per-request settlement.
Pricing that stays out of the way
We make money when you add credits — never on your tokens.
No per-token markup. You pay exactly the provider or seller price the request was routed to.
A flat 5% fee when you top up credits. That is the whole business model — no subscriptions, no seats.