543Models live

Every AI model, one key,
dirty cheap

One key for every model, priced below list — and you never pay for a request that fails.

One key for every major model

  • ImageOpenAI
  • ImageAnthropic
  • ImageGoogle
  • ImageKling
  • ImageByteDance
  • ImageRunway
  • ImageDeepSeek
  • ImagexAI
  • ImageAlibaba
  • ImageBlack Forest Labs
  • ImageMoonshot AI
  • ImageQwen
  • ImageMeta
  • ImageMistral
  • ImageMiniMax
  • ImageZhipu AI
  • ImagePixVerse
  • ImageLuma
  • ImageElevenLabs
  • ImageNVIDIA

Developer-first integration

One key.
Every kind of model.

Point the base URL at MartLLM and use the OpenAI or Anthropic SDK you already have. Nothing else in your code changes.

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.martllm.com/v1",
    api_key=os.environ["MARTLLM_API_KEY"],
)

reply = client.chat.completions.create(
    model="deepseek/deepseek-v4.1-flash-1017",
    messages=[{"role": "user", "content": "Hello"}],
)
print(reply.choices[0].message.content)

Reliability

Fast and reliable

Most models have many sellers. If one fails before answering, your request moves to the next one, and a failed request costs nothing.

  • A failed seller hands off to the next one
  • A failed request costs $0
  • Sellers get test requests every day
  • A fake seller is banned, its buyers refunded
How we check sellers
Paid requests, last 24 hours
284,476
Test requests sent to sellers, last 7 days
53,587
Models with 2 or more sellers
467 / 543

First token, median

  • ImageDeepSeek V4.1 Flash 10174.7 s
  • ImageClaude Haiku 5.51.7 s
  • ImageGPT-6 Luna2.6 s
  • ImageGoogle: Gemini 4 Hydrogen3.5 s

Privacy

Your data, your policies

Passwords and keys are swapped for tags before a request leaves MartLLM, and the real values come back in the answer. You choose the rest, key by key.

  • Passwords and API keys hidden, always
  • Emails, phones and cards hidden per key
  • Real values put back in every answer
  • Never used to train models
  • Requests not saved unless you ask
  • A header shows what was hidden
How we protect your data

API key settings

  • Hide passwords and API keysAlways on
  • Hide personal data from providers
  • Save my requests and answers

You send

Email jane@example.com
postgres://app:hunter2@db

The seller sees

Email [EMAIL_3f9a2c1b]
postgres://app:[SECRET_8d04e7a2]@db

Models live
543
Biggest saving off list price, right now
99.9%
On failed requests
$0

Models and current prices

See all models
Live · updated 10m ago

See all models →

Start building with the world’s top AI models today

MartLLM does not run models. Each request goes to a seller that serves the model through its own account with an API provider (for Claude, GPT and Gemini: the model maker, or a provider such as Amazon Bedrock, Snowflake, Venice AI or OpenRouter), at a price the seller sets below list. Sellers must own or be authorized to resell the access they list under their provider’s terms; stolen, leaked, shared, free-trial and evaluation keys are prohibited. A request that fails is not billed. How it works and why prices can be far below list

Developer-first integration

Up and running in minutes on one unified, OpenAI- and Anthropic-compatible API.

  • Pay less than list price
  • One key for every model
  • Routing that reroutes
  • Never pay for a failure

List capacity you may resell

Hold provider credits whose plan permits resale, at a provider on our approved list? List them on the market and earn when requests route to you. Another provider can be added on request, after we test it.

Frequently asked questions

Everything you need to know about integrating and billing with MartLLM.

Read the docs →
What is MartLLM?

MartLLM is one OpenAI- and Anthropic-compatible API key for Claude, GPT, Gemini, DeepSeek and other text, image, video and music models, usually below list price. It is an open market: sellers list offers for each model, each request goes to the cheapest working offer, and you pay what that seller charged.

How is it cheaper than going direct?

Sellers set their own prices as a discount off the model’s list price, and cheaper offers that are working get more of the requests. Sellers must have the right to resell what they list under their provider’s terms. The live price per model is in the pricing table; how it works is at martllm.com/how-it-works.

Which models can I use?

Leading LLMs (Claude, GPT, Gemini, DeepSeek and others) alongside image, video and music models. The market table lists every model that is live right now with its current price.

Do I pay for failed requests?

No. A request or a generation that fails costs $0. You are only billed for results you actually receive.

Is there a fee, a monthly fee or a minimum?

No MartLLM fee on requests. Adding credit by card costs 5.5% + $0.50. There is no monthly fee and no minimum spend: add credit from $5, keep one balance and pay as you go. Credit does not expire while your account is open.

Is it OpenAI-compatible?

Yes. Set the base URL to our endpoint and keep your existing OpenAI SDK; the Anthropic-compatible endpoint works the same way. You switch models by changing the model id.

What happens if a seller goes offline?

Your request is routed to the next healthy seller for that model automatically. For image and video jobs, task tracking and webhook callbacks mean an interruption on one seller doesn’t lose your job.

How do you handle my prompts and data?

Keys are scoped and revocable, and requests are encrypted in transit. Your traffic is used to serve your requests — not to train models.

How do I get started?

Create a free account, add credit (from $5), generate a key, point your SDK’s base URL at MartLLM, and send your first request. New accounts start at $0, so add credit before your first call.

Is MartLLM the same as UberLLM?

Yes. MartLLM was called UberLLM until November 2026. Same service, same accounts and API keys; uberllm.dev and api.uberllm.dev keep working.

Does MartLLM check that I get the model I asked for?

Yes, for sellers who list on MartLLM. They can list only through a provider on our approved list, and the provider’s model must be the model listed. Every offer is tested and reviewed by our team before any buyer reaches it, and keeps being tested on live requests. A seller caught serving a different model is banned, and the buyers it served are refunded from its unpaid earnings. What the checks catch and what they can’t: martllm.com/verification.

Can I sell my own capacity?

Yes, if you have the right to resell it under your provider’s terms and that provider is on our approved list (OpenRouter, Groq, Together AI, Fireworks AI, DeepSeek, Amazon Bedrock and others). Another provider can be added on request, after we test it. Your own server can’t be listed. See martllm.com/sell.