Context windows: what the numbers actually mean

The basics

Context window = how many tokens a model can see in one call. 128K tokens is roughly a 200-page book. 1M is roughly a full codebase or a long financial quarter. The window matters less than you think — and more than you'd guess — depending on the task.

When size changes the product

• 1M context (e.g. plugsky-micro) — load the whole codebase, the whole contract, the whole audit log. No chunking, no retrieval, no "the model forgot."
• 256K (plugsky-max) — long documents plus images plus reasoning in one call.
• 128K (most mid-tier models) — covers chat, support, summarization, and most agent loops.

The real tradeoff

Longer context costs compute. Models with 1M windows are slower and more expensive per token than compact models. The winning pattern is not "use the biggest window" — it is sending only what matters.

Pattern: keep the short model for chat, the long-context model for analysis passes, and RAG for retrieval. This is exactly the multi-model setup one endpoint makes easy.

Agents and context

Agents fail most often from context loss — the model forgets the original goal. A 1M window removes the need for aggressive summarization of the conversation history. That is why long-context models are the quiet backbone of reliable agent systems.

FAQ

Is bigger context always better?

No. Cost and latency grow with context. Match the window to the task.

Which models have the biggest windows?

plugsky-micro and plugsky-longctx support 1M context. See the full reference.