SkyForge · playground
Try the models before you rent the GPUs.
Send a few free prompts against GLM 5.3 Flash or Qwen3.8 Flash Next, served with vLLM on a shared Forge instance. No account needed for the first prompt — sign in for a couple more, launch your own instance for everything after that.
Shared demo instance
Ask something.
First prompt is free with no account. Replies are capped and the demo assistant keeps answers short.
Shared demo instance · prompts and token counts are logged for abuse prevention — see the privacy policy
01After the playground
Your own endpoint is the real product.
The playground runs on a shared instance with a small prompt cap. Launching gives you the same models on your own hardware, behind a private OpenAI-compatible /v1 endpoint.
Private /v1 endpoint
Your model, your instance, your API key. Point any OpenAI-compatible SDK at it and existing code works.
Per-second billing
Billed from boot to terminate on the GPU shape the launcher enforces. No commitment, no tier gates.
No prompt logging
Traffic to your private endpoint is proxied straight through — prompts and completions are never written to our logs.
Full catalog
The playground serves two fast models; the catalog covers coding, reasoning, and long-context shapes across the full GPU range.