Start Building with TokenLabs

Production-ready inference endpoints powered by NVIDIA DGX Spark. OpenAI-compatible API, pay-per-token pricing.

🔍
-
Models
128K
Context
$0.006
From / 1M Input
99.9%
Uptime

Inference Endpoints

Select a model to view pricing, try the playground, or grab API code.

Loading models...

Platform

⚡
Performance Benchmarks
Throughput, latency, and pricing across all quantization configs on DGX Spark hardware.
View benchmarks →
🎯
Accuracy Evaluation
IFEval instruction-following accuracy for every model and quantization configuration.
View accuracy →
📊
Baseline Comparison
Side-by-side comparison against cloud inference providers and baseline configurations.
View comparisons →
Overview
Playground
API Reference
Model Specifications
Pricing · per token
Input tokens
per 1M tokens
$0.00
Cached input tokens
per 1M tokens
$0.0000
Output tokens
per 1M tokens
$0.00

Prices derived from CI benchmarks (vLLM on DGX Spark). Free tier available.

Quick links
⚡Performance Benchmarks→ 🎯Accuracy Evaluation (IFEval)→ 📊Baseline Comparison→
API base URL

Compatible with openai, litellm, langchain.

All Available Models
Model
Primary
System Prompt
Parameters
Temperature0.70
Max Tokens512
Top P1.00
Options
Streaming
Mode: Select a model to compare
💬
Start a conversation
Send a message to chat with the selected model.
Quantization benefits
LLM haiku
FP8 vs NVFP4
About DGX Spark
Powered by vLLM · OpenAI-compatible API
Endpoint
POST

OpenAI-compatible chat completions. Streaming via stream: true.

Model identifier
Code Examples
cURL
Python
Node.js
LiteLLM

          

          

          

        

      

      

        
Request Parameters

        

          

            

              Parameter
              Type
              Description
            

            
              
modelstringFull model identifier

              
messagesarrayArray of {role, content} objects

              
max_tokensintegerMax tokens to generate (default 512)

              
temperaturefloatSampling temperature 0–2 (default 0.7)

              
top_pfloatNucleus sampling (default 1.0)

              
streambooleanStream via SSE (default false)