docs

LLMPad turns AI model popularity into an asset. Every major model family (Claude, GPT, Gemini, DeepSeek, Grok, …) can have exactly one official token on pump.fun. People stake that token and earn LLM credits from the token's creator rewards. Credits are spent on real AI usage through one API key — chat, image, and video models — or sold for SOL.

Two ways to earn: stake a model token and earn from its trading volume, or stake $LLM (the platform token) and earn from platform-wide distributions.

LLM credits are dollars of AI usage. $1 of credit pays for $1 of model usage at the provider's list price. Credits live in your balance and are spent through your API key.

Credits come in two kinds:

• Model credits — earned by staking a model's token. They only work on that model's family (CLAUDE credits → anthropic/* models).
• Universal credits — from $LLM staking, market buys, and the $1 welcome bonus. They work on every model, including image and video.

When you make a request, the model's own credits are spent first, then universal. If neither covers it, the request is refused — you can never go into debt.

Each model family can have one token, ever. First come, first served — once CLAUDE is launched, nobody else can launch another Claude token. The token's name, ticker, and image are the model's official identity and can't be customized.

Tokens launch on pump.fun through the launch page. Every token gets a page at llmpad.tech/m/$TICKER with live market data, staking, and activity.

Stake a model's token on its page. While staked, you earn a share of the LLM credits the token generates — proportional to your share of total staked. More trading volume on the token means more rewards; more total staked means your share is diluted.

• Staking is a real on-chain transfer — your tokens move to the pool's vault.
• Unstaking sends the tokens back to your wallet, minus nothing. No lockups, no cooldown.
• Earned credits accumulate in real time. Hit [ claim ] to move them to your spendable balance.
• APR on each page is estimated from recent rewards vs. the dollar value staked. It moves with volume.

$LLM is the platform's own token. Staking it earns universal credits — spendable on any model. Distributions to $LLM stakers are announced and split by stake share. Everything lives on the $LLM page.

One key, every model. Generate a key on the API page — it's shown once, so copy it. Then call the OpenAI-compatible endpoint:

curl https://llmpad.tech/api/public/v1/chat/completions \
  -H "Authorization: Bearer sk-ai-..." \
  -H "Content-Type: application/json" \
  -d '{"model": "anthropic/claude-opus-4", "messages": [{"role": "user", "content": "hi"}]}'

• Any OpenRouter model ID works — openai/gpt-5, google/gemini-2.5-pro, etc.
• You're billed the exact token cost of each call, at list price.
• No balance → the request is refused with a 402.
• Keys can be restricted to specific models and given spending limits on the API page.
• Streaming is not supported yet.

The API page also has a built-in try-it box so you can test without writing code.

Each key has a cost saver mode, set on the API page. Your code doesn't change — keep calling the same endpoint with the same model, and we route it cheaper:

• cheapest provider — same model you asked for, served by whichever provider charges the least. Quality is identical.
• cheapest model — we swap in the cheapest model of the same family (e.g. a Claude request may run on Haiku). Biggest savings, slightly lighter answers.
• off — exact model, exact provider.

Every response tells you which model actually ran and how much you saved versus the model you requested. Model-token credits stay locked to their family, so cheapest-model mode only swaps within that family.

The same key also generates images and video through Higgsfield models (Soul, DoP, Veo, Seedance, and more). Media jobs are priced per generation, charged up front from universal credits, and automatically refunded in full if the job fails or is rejected.

# start a job
curl -X POST https://llmpad.tech/api/public/v1/media/higgsfield-ai/soul/v2/standard \
  -H "Authorization: Bearer sk-ai-..." \
  -H "Content-Type: application/json" \
  -d '{"prompt": "a neon toad", ...}'

# check the result
curl https://llmpad.tech/api/public/v1/media/requests/<id> \
  -H "Authorization: Bearer sk-ai-..."

Video takes a minute or two — start the job, then poll the request URL until it's complete. The API page's [media] tab does this for you in the browser.

The market sells universal credits for SOL at 40% below list price — the cheapest way to use any model anywhere. Send SOL, credits land in your balance instantly. Selling credits back is not open yet.

Pick an unlaunched model family on the launch page, pay the launch fee (plus an optional first buy), and the token goes live on pump.fun with its official page on LLMPad. A model is only reserved once the launch is paid — unpaid attempts don't block anyone.

Trading agents are SOL-funded bots that trade memecoins and pay profits out as LLM credits to their owner. New agent creation is currently paused. Existing agents can be withdrawn at any time from the agents page.

Are the credits real money? They're real AI usage. Spend them on the API, or (when selling opens) sell them for SOL.

Do I need an account? Just a Solana wallet. Connecting signs a message — no transaction, no fees, no email.

Can I unstake anytime? Yes. Tokens go straight back to your wallet.

I staked CLAUDE — can I only use Claude? Your CLAUDE credits, yes. But universal credits (welcome bonus, $LLM, market buys) work on everything.

Lost my API key? Keys are stored hashed — we can't show it again. Generate a new one on the API page.