No-KYC LLM API: no email, no ID, open models paid in crypto

A WeightsAPI account takes a username of 3 to 32 characters and a password of at least 15 characters. No email address, phone number or identity document is requested or stored, and there is no card field: the balance is funded in crypto only.

Signup shows a one-time recovery code. With no email on file, it is the only way to reset a lost password, so keep it offline next to your wallet backup.

No KYC is not anonymity: keys, usage and payments are recorded against your account, which is how balances and spending caps work without knowing who you are.

The account keeps what billing needs: key metadata (name, visible prefix, SHA-256 hash, cap, allowed models), token totals, charges, and payment records (order reference, asset, network, receiving address, USD target, transaction details). The trust and security page lists every record type; retention periods are not published yet.

Prompt text is not on that list: the gateway does not persist prompts, completions or tool arguments, and its logs omit request bodies. Upstream provider retention is not yet verified, so this is not a zero-retention promise; data handling shows exactly where the gateway’s guarantee stops.

A transfer that arrives after the 10-minute quote window, or for a different amount, goes to review. A public transaction reference you share with support can link activity to a wallet.

Every top-up lands in one USD balance, shared by the playground and the API at https://weightsapi.com/v1. Rates per million tokens, input / output: Llama 3.1 8B $0.03 / $0.05, Gemma 3 27B $0.10 / $0.15, Qwen3 32B $0.15 / $0.20, Hermes 4 70B $0.13 / $0.40 and Dolphin Mistral 24B Venice Edition $0.20 / $0.90.

The same balance prepays one month of a dedicated GPU: L40S (48 GB) $416, A100 (80 GB) $792, H100 SXM (80 GB) $1,352, H200 (141 GB) $1,832, 2 × H100 $2,568 and 8 × H100 $9,736. The month starts on activation and never renews by itself. By comparison, RunPod lists an H100 SXM pod at $3.49 per hour as of 25 September 2026: about $2,548 over a 730-hour month (our calculation), so a machine that runs around the clock costs about $1,196 less per month here.

None of the OpenRouter, Venice or NanoGPT pages cited below offers a monthly dedicated GPU, and RunPod and Together AI price GPUs by the hour. Here, variable traffic runs per token and steady load moves to a flat monthly machine, on one account.

1. Nothing personal at signup

Username, password, recovery code: no inbox to verify, no document to upload, no card to link. Compare: OpenAI’s docs say its latest reasoning models may require organization verification, which TechCrunch reported in April 2025 uses a government-issued ID, and RunPod asks for any required KYC before a first crypto payment. Create a username account.

2. Monero, Bitcoin or stablecoins, with no platform deposit fee

Pay in BTC, ETH, USDT on TRON or Ethereum, USDC on Ethereum, XMR, SOL, LTC or TRX. The amount is fixed for 10 minutes at a recent market rate, and a USD 100 top-up credits USD 100 within about a minute once the network confirmations are reached: you pay only the network fee. OpenRouter charges 5% on crypto credit purchases and accepts USDC, so a USD 1,000 top-up there carries a $50 fee that you do not pay here. See every asset and confirmation count.

3. Tokens and a monthly GPU on one balance

When a workload outgrows shared inference, you skip a second provider, a second identity check and a second payment rail. An H100 SXM at $1,352 for one month from activation is a flat figure you can budget. Compare the six GPU configurations.

4. Lower token prices where they matter

We are not cheapest on every model; here is where we beat OpenRouter’s listed rates as of 25 September 2026: Mixtral 8×22B $0.60 / $0.60 versus $2.00 / $6.00, Gemma 3 27B output $0.15 versus $0.45, Llama 3.1 8B $0.03 / $0.05 versus $0.05 / $0.08, DeepSeek V3 output $0.85 versus $1.00, Qwen3 32B output $0.20 versus $0.28. On Llama 3.3 70B, OpenRouter is cheaper ($0.10 / $0.32 versus our $0.35 / $0.40). See the model-by-model comparison.

5. Key controls instead of identity checks

Each API key carries a lifetime USD spending cap, an optional model allowlist and revocation. Give every tool or SillyTavern setup its own capped key, and a leaked key can only spend what you allowed. Read how keys work.

6. No content filter added at the gateway

We add no filter between your request and the model. Fine-tunes such as Dolphin Mistral 24B Venice Edition and Hermes 4 70B answer as their publishers trained them; a model’s own training can still refuse. See models for creative and roleplay work.

Full rates are on the pricing page, and asset rules are in billing and crypto deposits.

Rows use each provider’s own public pages.

Where others fit better: NanoGPT needs no account and takes deposits from $0.10, and Venice offers USDC wallet access without an account. For tiny top-ups, use them. For no email, no ID, Monero without a platform fee and a monthly GPU on the same balance, choose WeightsAPI.

What will it really cost? Published token rates plus your network fee; with no platform deposit fee, the whole top-up becomes credit. USD 20 is a per-top-up minimum, not a balance floor: a balance below $20 keeps paying for requests, and fewer, larger transfers mean fewer network fees. The pricing page’s example, 100 million Qwen3 32B tokens a month at 80% input, costs $16.00, so a USD 100 top-up covers about six such months (our calculation).

Can I trust it with my money? Commit only what you will use: make your first top-up the USD 20 minimum and add more once your model is serving, since credit cannot be withdrawn. Rates are in prices.json, conditions in the terms, and the trust page states what is covered today and what is not yet published.

Is my model available? Check the status page, which lists all 15 models and refreshes every 60 seconds, before you top up. A GPU month starts only on the activation date the operator confirms.

How do I pay without mistakes? Pick the asset and network at checkout, send the exact quoted amount within 10 minutes and keep the order reference. A transfer on another network has no established recovery route.

How long until credit appears? Your balance is credited within about a minute once the network confirmations are reached: 1 for BTC, 6 for LTC, 10 for XMR, 12 for ETH, USDC and USDT on Ethereum, 19 for USDT on TRON and TRX, 32 for SOL. You can start as soon as your transfer confirms; a pending payment does not unlock access.

Will anyone ask for ID later? Not in the current account flow. What remains are the account records and the review of late or mismatched transfers described above.

What are the GPU conditions? One month from activation, no automatic renewal; cancelling before activation refunds to your balance. Shell access, backups, failover and autoscaling are not included by default, and the operator confirms model fit, region and timing per order.

Choose WeightsAPI if you want an OpenAI-compatible API without an email, ID or card; you pay in crypto and refuse a percentage deposit fee; you run Mixtral 8×22B, Gemma 3 27B, Llama 3.1 8B, DeepSeek V3 or output-heavy Qwen3 32B; or you plan to move steady traffic to a monthly GPU.

Look elsewhere if you need top-ups under USD 20, no account at all, closed models such as GPT or Claude, a contractual SLA, or the lowest Llama 3.3 70B price.

What to do now: create a username account and store the recovery code offline, check your model on the status page, then top up from USD 20 in the asset you already hold. Planning a GPU? Choose a configuration and get the activation date confirmed before you pay.