Context Windows, Compared: ChatGPT vs Claude vs Gemini

The “context window” is an AI LLM’s working memory: the total number of tokens — your prompts, its replies, pasted documents, everything — the model can hold at once. When the conversation exceeds it, the oldest content falls out of view. To the model, those turns never happened.

The state of play

The three major LLM families all offer large windows, and the exact numbers change with every release and subscription tier:

• ChatGPT (OpenAI) — flagship models offer very large contexts; smaller, faster models get smaller budgets.
• Claude (Anthropic) — known for especially large windows on top models, with tier-dependent limits.
• Gemini (Google) — competitive flagship windows, also tier-dependent.

Exact figures move fast enough that any number printed in a blog post is stale by the time you read it — check each vendor’s current docs. The shape of the problem is stable, though: the ceiling is real, it counts everything, and it’s measured in tokens.

Why you hit the ceiling mid-project

Everything shares one budget: system prompts, conversation history, pasted documents, and the model’s own answers. A long-running project chat slowly fills the window with turns that add nothing new — and the day the window overflows, the model forgets the constraints you set at the top.

Living inside the window

1. Start fresh chats for new tasks instead of piling onto one thread.
2. Compress before you paste. A trimmed document or system prompt is a smaller occupant of the same window — and unlike a rate, it’s fully in your control.
3. Keep durable rules in a file you re-paste, rather than trusting old turns to survive.
4. Watch your count. An estimator (like the one in our compressor) tells you when a paste is about to get expensive in window terms.

Compress your long prompts at supatoken.lol — free, instant, and nothing is stored.