Claude first · pre-launch

Kudos for
Claude

Claude is the first model on Kudos. Hold $KUDOS, earn OpenRouter credits every hour, and spend them on Opus, Sonnet and Haiku wherever you already work.

How the payout works
What your credits buy

Three Claude models,
one key.

Frontier

Claude Opus 5

The heavy model for long agentic runs and hard reasoning. Where most of a Claude Code session's spend goes.

anthropic/claude-opus-5
Daily driver

Claude Sonnet 5

Fast enough for editor loops, strong enough to ship. The default we set in every setup guide below.

anthropic/claude-sonnet-5
Volume

Claude Haiku 4.5

Cheap, quick calls for classification, extraction and sub-agents. Stretches a small balance a long way.

anthropic/claude-haiku-4.5
Prices on the credits marketplace track today's best OpenRouter ask, 30% off. Requests go from your machine straight to OpenRouter with your key. We never proxy the traffic and never see a prompt.
Setup

Paste the key,
keep working.

What you claim is a plain OpenRouter key. Claude Code speaks to OpenRouter natively, Cursor accepts it as an OpenAI-compatible endpoint, and the Anthropic SDK just needs a base URL. Nothing to install.

Keys are issued one tranche at a time, up to $200. When one is spent, claim again for a fresh one and swap it into whichever setup you use.

Set the gateway variables and launch. No proxy in between.

terminal 
export ANTHROPIC_BASE_URL="https://openrouter.ai/api"
export ANTHROPIC_AUTH_TOKEN="sk-or-v1-…"  # the key you claimed
export ANTHROPIC_API_KEY=""                # must stay empty
claude
Signed in with an Anthropic account before? Run /logout once and relaunch, then confirm with /status that the base URL points at OpenRouter.

Paste the key under Models, override the base URL, and pick a Claude model.

cursor · settings › models 
OpenAI API Key      sk-or-v1-…        # the key you claimed
Override base URL   https://openrouter.ai/api/v1
Model               anthropic/claude-sonnet-5
Add the model id by hand under "Add model" if it is not in the dropdown, then hit the test button before closing settings.

The Anthropic SDK with a base URL swap. Every other OpenAI-compatible SDK works the same way.

python 
from anthropic import Anthropic

client = Anthropic(
    base_url="https://openrouter.ai/api",
    auth_token="sk-or-v1-…",  # the key you claimed
)
msg = client.messages.create(
    model="anthropic/claude-sonnet-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": "hi"}],
)
Streaming, tool use and prompt caching all pass through unchanged.
Coming next

One model at a time.

We are switching models on one by one, with setup guides and marketplace pricing checked for each before it goes live. Claude is first because it is where Claude Code and most agent spend already lives.

GPT soonGemini soonDeepSeek soonKimi soonGrok soon
Questions
Why Claude first?
Because the people holding $KUDOS are mostly running Claude Code, and that is where an inference balance is spent fastest. Supporting one model well beats listing three hundred we have not tested.
Can the key call other models?
The key is a standard OpenRouter key and we never proxy or block traffic, so nothing technical stops it. What Kudos supports, documents and prices on the marketplace is Claude today. Other models are unsupported until we switch them on, and you will not need a new key when we do.
Which Claude models exactly?
Opus 5, Sonnet 5 and Haiku 4.5 through OpenRouter's anthropic routes. When Anthropic ships a new model, it appears on the marketplace as soon as OpenRouter routes it.
When do the others arrive?
After the Claude flow has cleared smoothly on testnet and mainnet. Each new model gets its own setup guide and pricing check first. Follow @KudosLLM for the switch-ons.