One API for every model you'll ever need.

What is an inference API?

An inference API lets your application send prompts to a hosted model and receive completions over HTTP — no GPUs, no model management, no infrastructure. Plugsky's inference API is OpenAI-compatible, so your existing OpenAI code works by changing the base URL.

What you get

• Chat completions — text, code, reasoning, and long-context tasks.
• Streaming — token-by-token responses over SSE for fast UX.
• Embeddings — 2,048- and 4,096-dimension vectors for RAG and search.
• Vision models — image understanding with multimodal models.
• Function calling — structured tool use for agents.

Models on the API

30+ models across tiers: compact models like plugsky-lite for speed, mid-size reasoning models, and frontier-tier models like plugsky-frontier (Mistral Large 3 675B, 128K context). One API key, one endpoint, every model.

Latency

Real measured latencies on the live API: plugsky-phi ~0.23s, plugsky-lite ~0.24s, plugsky-gemma-4 ~0.28s, plugsky-micro ~0.55s, plugsky-pro ~0.67s, plugsky-frontier ~0.70s, plugsky-max ~0.71s. Your mileage depends on model size and workload, but the API is built for speed.

Inference API FAQ

Can I use the OpenAI Python SDK?

Yes — set base_url to https://api.plugsky.com/v1 and your code works. See the quickstart.

What happens if a model is overloaded?

Requests fail over automatically to a backup upstream, so your call still returns instead of timing out.

Is there a free tier?

Yes — every plan includes access, and the Free tier is $0. See pricing.