Pay for an LLM API and GPU rental with crypto: BTC, XMR, USDT, USDC, ETH, SOL, LTC

Every confirmed top-up becomes USD credit in one balance. The playground and the API draw on it per million tokens, input and output priced separately, from $0.03 / $0.05 on Llama 3.1 8B to $1.20 / $1.20 on Llama 3.1 405B. No subscription; a request rejected before dispatch costs nothing.

The same balance prepays a dedicated GPU for one month: L40S (48 GB) $416, A100 (80 GB) $792, H100 SXM (80 GB) $1,352, H200 (141 GB) $1,832, 2 × H100 $2,568 or 8 × H100 $9,736. No per-token charges apply on that machine; the month starts on activation and does not renew automatically.

What USD 100 means in tokens, computed from our published rates at 80% input and 20% output: about 2.9 billion tokens on Llama 3.1 8B, 625 million on Qwen3 32B or 270 million on DeepSeek V3.

Checkout offers nine asset and network options. A transfer is credited within about a minute once it reaches its confirmation count, and USDT and USDC are quoted at their market rate, not an assumed $1.

Send only on the network shown on your funding order: USDT on TRON and USDT on Ethereum are not interchangeable. Full rules are in billing and crypto deposits.

1. Create an account. A username (3 to 32 characters) and a password of at least 15. Store the one-time recovery code offline: it is the only password recovery.
2. Open a funding order for USD 20 or more. Pick the asset and network. Checkout converts your USD amount at a recent CoinPaprika rate (Kraken as fallback) and fixes the crypto amount for 10 minutes.
3. Send the exact amount to the address on the order, on the displayed network, plus your wallet’s network fee. Keep the order reference and transaction ID.
4. Wait for the network confirmations. Your balance is credited within about a minute once they are reached, so you can start right away. A late or different amount goes to review, so never pay twice while a payment is pending.
5. Spend it. Create API keys with a lifetime spending cap and a model allowlist, or pay a GPU order from confirmed credit; a top-up never pays a rental automatically.

1. No platform fee between your wallet and your balance

An exact payment inside the quote window qualifies for the full USD amount on your order, credited within about a minute once the network confirmations are reached; your only extra cost is the network fee. OpenRouter adds a 5% fee to crypto credit purchases, about $5 per $100. With us that $5 stays in tokens or GPU time. Read the billing rules.

2. Pay in the coin you already hold, Monero included

Eight coins fund the same balance, so there is no swap step. OpenRouter takes USDC for crypto payments, and Venice sells API credit in USD or USDC. Holding XMR or USDT on TRON? Pay us directly. Check your coin and its confirmation count.

3. Tokens first, a monthly GPU later, same balance

Prototype on the shared API, then move a steady workload to a dedicated GPU without a second account, check or provider. An H100 SXM month is $1,352, against about $2,548 for 730 hours at RunPod’s $3.49 Secure Cloud rate, and RunPod’s billing docs ask for any required KYC before a first crypto payment. Compare the six GPU configurations.

4. No KYC step, and a straight answer on what is recorded

Signup is a username and a password: no email, phone, ID document or card. No KYC is not anonymity: keys, usage and payment records (asset, network, address, amount, transaction details) stay linked to your account. See exactly what signup asks for.

5. Published prices, including where we lose

Every rate is in prices.json. We beat OpenRouter on Mixtral 8×22B and Llama 3.1 8B, and on output for DeepSeek V3, Gemma 3 27B and Qwen3 32B; OpenRouter is cheaper on Llama 3.3 70B and DeepSeek V4 Flash. See the model-by-model table.

All rates are on the pricing page.

Where another option fits better: NanoGPT to start with cents and no account; Venice for USDC wallet access without an account; RunPod for a GPU used under roughly 380 to 400 hours a month on Secure Cloud, or about 500 to 530 on Community Cloud, where per-second billing costs less.

Same model, two prices

OpenRouter rates exclude its 5% crypto fee. On Llama 3.3 70B, Venice charges $0.70 / $2.80 and Together AI $1.04 / $1.04. More rows: our OpenRouter alternative comparison.

Monthly GPU versus hourly GPU

RunPod’s Community Cloud is cheaper ($0.79, $2.69 and $3.59 an hour), which moves break-even to about 527, 503 and 510 hours (computed). Not like-for-like: our month is managed inference capacity, with no shell access, backups, failover or autoscaling by default, while RunPod bills per second.

We do not sell GPT or Claude. We sell open-weight models behind an OpenAI-compatible endpoint: set the base URL to https://weightsapi.com/v1 with a WeightsAPI key and an exact model ID. Chat Completions, Completions and the model list are covered; the Responses and Assistants APIs are not drop-in. Check the migration guide.

Open-weight picks by job, in USD per 1M input / output tokens. They are different models from GPT and Claude, so test them on your own prompts:

• General chat: Llama 3.3 70B ($0.35 / $0.40) or Qwen3 235B A22B ($0.20 / $0.60)
• Reasoning: DeepSeek R1 ($0.50 / $2.15)
• Coding: Qwen3-Coder-Next ($0.12 / $0.80)
• Image input: Gemma 3 27B ($0.10 / $0.15)
• High volume: Llama 3.1 8B ($0.03 / $0.05) or DeepSeek V4 Flash ($0.09 / $0.18)
• Creative work and roleplay: Dolphin Mistral 24B Venice Edition ($0.20 / $0.90) and Hermes 4 70B ($0.13 / $0.40), with no content filter added at the gateway

What will it really cost? Your top-up plus the network fee, then tokens at published rates (reserved for input plus maximum output, settled on confirmed usage) or a flat GPU month. A spending cap on each key stops a runaway script.

Can I trust it with my money? Rates, terms and the records we keep (on the trust page) are published, gaps included; no SLA is offered. Start with one USD 20 top-up and scale from there.

Is my model available? Shared inference is in launch preview: check your model on the status page (refreshed every 60 seconds) before topping up. For GPUs, the activation date is confirmed before you pay.

Will anyone ask for ID? Not in the current account flow, for tokens or GPUs. A mismatched transfer is reviewed, and a support report may include a transaction ID, which can link activity to a wallet.

What are the conditions? Credit cannot be withdrawn to a wallet; cancelling a GPU order before activation returns the payment to your balance (refund FAQ). A GPU month does not renew automatically.

Choose WeightsAPI if you hold BTC, XMR, USDT or another listed coin and want open-weight inference and GPU capacity without a platform fee, a card or an ID check, especially if your usage will grow into a monthly GPU or leans on the models where our rate is lower.

Look elsewhere if you want to test with less than USD 20, need GPT or Claude themselves, need an SLA, or need unused crypto returned to a wallet.

What to do now: create a username account, check your model on /status, then top up from USD 20 in the coin you hold. For a GPU, pick a configuration on dedicated GPUs and get the activation date confirmed before you pay.