Open models.
An API you already know.
Token Factory serves open-weight models behind an OpenAI-compatible endpoint, billed per token. Change the base URL, keep your code, and never size a GPU.
import os from openai import OpenAI client = OpenAI( base_url="https://api.packet.ai/v1", api_key=os.environ["PACKET_API_KEY"], ) resp = client.chat.completions.create( model="meta-llama/Llama-3.3-70B-Instruct", messages=[{"role": "user", "content": "Hi!"}], stream=True, )
Inference without
the infrastructure.
For apps whose traffic doesn't justify a GPU running all day.
Drop-in compatible
Chat completions with streaming, JSON mode and tool calling, through the OpenAI SDKs you already use.
Pay per token
Input and output tokens metered separately from the same wallet as your GPUs. Nothing to pay while idle.
Graduate to a GPU
When volume makes a dedicated deployment cheaper, serve the same model on your own GPU with vLLM, same endpoint shape.
Three models live.
One endpoint.
Free at launch (one month), then billed per 1M tokens.
DeepSeek V4.1 Flash NVFP4
A reasoning model with a 1M-token window and image input, in NVIDIA's 4-bit build.
- context
- 1M
- input
- Vision
- license
- MIT
nvidia/DeepSeek-V4.1-Flash-NVFP4View model MiniMax M3.1 Flash
An agentic model for coding and computer use, with text, image and video input.
- context
- 1M
- input
- Vision + video
- reasoning effort
- 5 levels
MiniMaxAI/MiniMax-M3.1-FlashView model Qwen2.5 1.5B Instruct
A small, quick model for classification, extraction and other high-volume jobs.
- context
- 32K
- languages
- 29+
- license
- Apache 2.0
Qwen/Qwen2.5-1.5B-InstructView model Your next run
starts at $0.36.
Sign up, top up, launch. You'll be on a GPU in minutes.