Context Window Budget Calculator
See exactly how much of your model's context window your system prompt, tool definitions, and conversation history are eating — before you find out mid-conversation.
0 tokens
0 tokens
0 tokens
Tokens set aside for the model's answer
Budget breakdown
GPT-5.6 Luna — 1,050,000 token window1,045,904 tokens left for the actual task
99.6% of the context window is still free
Why context budgeting matters more than raw context size
A 1M-token context window sounds enormous until a long system prompt, a dozen registered MCP tools, and a growing conversation history quietly eat most of it. Agentic setups are the worst offenders — every tool definition you register (see the MCP Tool Generator) is tokens spent before the model has read a single word of the actual task.
This is the practical side of what the industry now calls "context engineering" — treating context window space as a real, finite budget to manage deliberately, rather than an afterthought.
Frequently asked questions
How accurate is the token estimate?
It uses the same ~4 characters ≈ 1 token approximation as our Token Calculator, which is the standard rough estimate for English text used across providers. Actual counts vary by tokenizer and language.
Where do the context window sizes come from?
Provider documentation and OpenRouter model listings as of August 2026. This market moves fast — always confirm the exact figure on your provider's pricing/docs page before relying on it for production limits.
Why reserve tokens for the response?
Most providers count your output against the same context window as your input. If you don't reserve space for it, a long response can get cut off even though your prompt technically "fit."
Is my data stored?
No. All calculation happens in your browser — nothing is uploaded.