No KYC
No identity-document upload or KYC step in the current sign-in flow. Access and usage remain linked to your account.
Navigation
Managed LLM inference. Open models, one API.
Choose published open-weight models through an OpenAI-compatible API. Keep familiar tools, compare token costs and match the model to your workload.
No added content filters
WeightsAPI adds no content filter at the gateway. A model’s own training can still lead it to refuse a request.
Shape your first request.
Choose a catalogue model, write a small prompt and inspect the matching code. Keep the first example simple so the request and expected answer are easy to check.
Account access Sign in and confirm a top-up of at least USD 100, then spend your prepaid credit. Every recharge has the same minimum; smaller remaining balances stay usable. Check service status for model availability.
PlaygroundFunded account required
Checking account access…
Prompt content is not saved by this gateway.
ResponseReady when account, key and service checks pass
Check access before your first request
Sign in, confirm a top-up of at least USD 100 and select an API key to send your first request.curl https://weightsapi.com/v1/chat/completions \
-H "Authorization: Bearer $WEIGHTSAPI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "Qwen/Qwen3-32B",
"messages": [
{
"role": "user",
"content": "Hello!"
}
],
"stream": true,
"max_tokens": 512
}'The same request, in your application.Read the quickstart
* Service context. See the model page for the native window and configuration requirements.
Build around your application
Start with a useful task.
The cookbook gives you small examples to adapt, with the application still in control.
Conversational assistants
Build a conversation loop, manage message history and validate the answers your users receive. Start with a terminal example before adding a user interface.
Coding and tools
Explore a tool-calling loop with an explicit allowlist and argument checks. Model and deployment support must be verified before enabling actions.
Answers from documents
Bring your own retriever, supply selected passages and verify the citations. The example uses simple text matching; the embeddings service is not available.
A practical starting point
Plan your first workflow.
Decide what your application needs before increasing traffic or adding more features.
- Compare models
Choose for the task
Compare capabilities, model licenses, context and input/output rates. Check service status and the confirmed deployment limits before routing requests to a model.
- Open the quickstart
Prepare a small request
Use the chat-completions format with an exact model ID and a bounded output. Read the streaming and error guides before adding retries or tools.
- Explore the console
Keep usage understandable
Separate applications with distinct keys and spending caps. The console is the place to review usage and available credit, including amounts reserved for pending requests.
Model catalogue
Choose the right tradeoff.
Compare the published choices below, then open a model page for its capabilities, license and deployment notes. Price alone does not establish quality or speed.
| Model | Input | Output | Service context |
|---|---|---|---|
| Llama 3.1 8B | $0.03 | $0.05 | 131,072 |
| Qwen3 32B | $0.15 | $0.20 | 131,072 |
| DeepSeek V3 | $0.25 | $0.85 | 131,072 |
| DeepSeek R1 | $0.50 | $2.15 | 131,072 |
Input and output rates are in USD per million tokens. Service context limits require deployment confirmation; model-native limits are separate.
Your existing tools
Bring your existing tools.
Connect familiar tools using their documented settings. Validate the features your application needs on the selected deployment before moving traffic.
SillyTavern
Configure a custom OpenAI-compatible chat connection and check the model selection before sending a test message.
Open WebUI
Set up an OpenAI API connection for a chat interface, with model discovery or an explicit model filter.
Cline
Prepare the OpenAI Compatible provider and align context, output limits and capability settings with the actual deployment.
n8n
Start with an OpenAI Chat Model in a basic chain. Agent workflows need separately verified tool-calling support.
Know the boundaries
Understand the service boundaries.
Review account controls, data handling and deployment requirements before choosing a model.
A defined data boundary
The gateway does not persist prompts or responses. Provider retention and infrastructure logging still need verification; this is not an end-to-end retention guarantee.
Rates you can inspect
Compare input and output separately and estimate with your own workload. Dedicated capacity follows a different reservation model.
A path through problems
Use the support guide to diagnose errors and prepare a report. Check service status for availability and monitoring information.
Before you begin
Questions before you begin.
Understand availability, compatibility, costs and data handling before building them into your application.
Do I need to complete KYC?
The current sign-in flow has no identity-document upload or KYC verification step. A signed-in account is still required, and usage remains associated with that account.
Are the models uncensored?
WeightsAPI adds no content filter at the gateway. The catalogue includes original models and fine-tuned variants published by their respective creators, including Dolphin and Hermes. Some variants are designed to reduce refusals; model training can still cause refusals. Read each model’s source and behavior notes.
What can I use today?
Manage API keys, create crypto funding orders and place dedicated GPU orders. Check service status for model availability; payment credit and GPU activation require confirmation.
What does OpenAI-compatible mean here?
The API uses Chat Completions, legacy Completions and model discovery routes. Check the selected deployment’s capabilities; provider-specific APIs are not interchangeable.
How is a model request priced?
Costs use separate model rates for actual input and output tokens, settled from provider-confirmed usage after an initial credit reservation.
Does WeightsAPI store my prompts or model responses?
The gateway does not persist prompts or responses in its application database. Account records remain, and provider retention requires separate verification.
Choose a model. Make the first request clear.
Start with the catalogue and a small example. The docs explain the request format; the FAQ helps resolve the decisions around it.