Tendril

Tendril · Builders · Model Families

Context Windows: How Much AI Can 'Remember'

Each AI has a 'context window' — how much it can hold in memory. Knowing this matters for big tasks.

40 min · Reviewed 2026

The big idea

Each AI can only 'remember' a certain amount of text in one chat. This is the context window. Bigger window = can handle longer documents. Understanding this saves you from confusion.

Some examples

GPT-4 / Claude: huge context windows now (millions of tokens).
Smaller models: smaller windows.
If your chat gets very long, AI may forget early stuff.
For long documents, paste the whole thing at the START of the chat.

Try it!

Understanding "Context Windows: How Much AI Can 'Remember'" in practice: Understanding AI in this area gives you a real advantage in how you work and think. Each AI has a 'context window' — how much it can hold in memory. Knowing this matters for big tasks — and knowing how to apply this gives you a concrete advantage.

Apply context window in your model-families workflow to get better results
Apply AI memory in your model-families workflow to get better results
Apply limits in your model-families workflow to get better results

Apply Context Windows: How Much AI Can 'Remember' in a live project this week
Write a short summary of what you'd do differently after learning this
Share one insight with a colleague

Why Gemini's 2M Context Window Matters (and Doesn't)

The big idea

Context window = how much text you can paste in one go. Claude offers 200K (or 1M for some). GPT-5 offers 400K. Gemini offers 2M (about a Lord of the Rings book). Sounds amazing — but research shows models get worse at finding info in the middle of long contexts ('lost in the middle'). More context isn't always better quality.

Some examples

Pasting a 500-page PDF and asking 'what does it say?' — surface info works fine.
Asking 'find the one sentence about taxes on page 213' — performance drops noticeably.
For long docs, splitting into chunks and asking targeted questions often beats one big paste.
RAG (retrieval-augmented generation) tools like NotebookLM are designed exactly for this.

Try it!

Take a long article (10+ pages). Paste into Gemini and ask a specific question. Then split it into 3 parts and ask the same question. Compare answers.

Why Some Models Have 1 Million-Token Context Windows

The big idea

The context window is how much text the model can read at once. GPT-3 had 4,000 tokens (a few pages). Claude 4.5 has 200k+ (a small book). Gemini 2.5 has 1-2 million (a whole textbook). Bigger context = more material the AI can reason over without needing summaries.

Some examples

You drop a 300-page novel into Gemini and ask 'what foreshadowing did chapter 3 contain about chapter 27?'
You paste a whole codebase into Claude and ask 'where would I add a feature flag?'
You paste 30 PDFs of research into Gemini and ask for a meta-analysis.
You paste a 2-hour transcript of a podcast and ask for chapter timestamps.

Try it!

Find a long document (a novel, a codebase, a course textbook). Paste as much as you can into Gemini or Claude. Ask a question that requires connecting two distant parts.

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-builders-model-families-AI-and-context-windows-teen

What is the core idea behind "Context Windows: How Much AI Can 'Remember'"?
1. Each AI has a 'context window' — how much it can hold in memory. Knowing this matters for big tasks.
2. Behave identically across families on the same prompt
3. Use for: cost-sensitive use cases at high token volume (Gemini Flash is competit…
4. Measure cache hit rates per provider
Which term best describes a foundational idea in "Context Windows: How Much AI Can 'Remember'"?
1. AI memory
2. context window
3. limits
4. Behave identically across families on the same prompt
A learner studying Context Windows: How Much AI Can 'Remember' would need to understand which concept?
1. context window
2. limits
3. AI memory
4. Behave identically across families on the same prompt
Which of these is directly relevant to Context Windows: How Much AI Can 'Remember'?
1. context window
2. AI memory
3. Behave identically across families on the same prompt
4. limits
Which of the following is a key point about Context Windows: How Much AI Can 'Remember'?
1. GPT-4 / Claude: huge context windows now (millions of tokens).
2. Smaller models: smaller windows.
3. If your chat gets very long, AI may forget early stuff.
4. For long documents, paste the whole thing at the START of the chat.
Which of these does NOT belong in a discussion of Context Windows: How Much AI Can 'Remember'?
1. Behave identically across families on the same prompt
2. Smaller models: smaller windows.
3. If your chat gets very long, AI may forget early stuff.
4. GPT-4 / Claude: huge context windows now (millions of tokens).
Which statement is accurate regarding Context Windows: How Much AI Can 'Remember'?
1. Apply AI memory in your model-families workflow to get better results
2. Apply limits in your model-families workflow to get better results
3. Apply context window in your model-families workflow to get better results
4. Behave identically across families on the same prompt
Which of these correctly reflects a principle in Context Windows: How Much AI Can 'Remember'?
1. Write a short summary of what you'd do differently after learning this
2. Share one insight with a colleague
3. Behave identically across families on the same prompt
4. Apply Context Windows: How Much AI Can 'Remember' in a live project this week
What is the key insight about "The rule" in the context of Context Windows: How Much AI Can 'Remember'?
1. Match the AI to the size of your task. Big task = pick AI with big context window.
2. Behave identically across families on the same prompt
3. Use for: cost-sensitive use cases at high token volume (Gemini Flash is competit…
4. Measure cache hit rates per provider
What is the recommended tip about "Match model to task" in the context of Context Windows: How Much AI Can 'Remember'?
1. Behave identically across families on the same prompt
2. Different models have different strengths. A coding model won't write great poetry; a creative model won't write tight S…
3. Use for: cost-sensitive use cases at high token volume (Gemini Flash is competit…
4. Measure cache hit rates per provider
Which statement accurately describes an aspect of Context Windows: How Much AI Can 'Remember'?
1. Behave identically across families on the same prompt
2. Use for: cost-sensitive use cases at high token volume (Gemini Flash is competit…
3. Each AI can only 'remember' a certain amount of text in one chat. This is the context window. Bigger window = can handle longer documents.
4. Measure cache hit rates per provider
What does working with Context Windows: How Much AI Can 'Remember' typically involve?
1. Behave identically across families on the same prompt
2. Use for: cost-sensitive use cases at high token volume (Gemini Flash is competit…
3. Measure cache hit rates per provider
4. Understanding "Context Windows: How Much AI Can 'Remember'" in practice: Understanding AI in this area gives you a real advantage in how you…
Which best describes the scope of "Context Windows: How Much AI Can 'Remember'"?
1. It focuses on Each AI has a 'context window' — how much it can hold in memory. Knowing this matters for big tasks.
2. It is unrelated to model-families workflows
3. It applies only to the opposite beginner tier
4. It was deprecated in 2024 and no longer relevant
Which section heading best belongs in a lesson about Context Windows: How Much AI Can 'Remember'?
1. Behave identically across families on the same prompt
2. Some examples
3. Use for: cost-sensitive use cases at high token volume (Gemini Flash is competit…
4. Measure cache hit rates per provider
Which section heading best belongs in a lesson about Context Windows: How Much AI Can 'Remember'?
1. Behave identically across families on the same prompt
2. Use for: cost-sensitive use cases at high token volume (Gemini Flash is competit…
3. Try it!
4. Measure cache hit rates per provider

← Back to interactive lesson

Tendril · Builders · Model Families

Context Windows: How Much AI Can 'Remember'

Each AI has a 'context window' — how much it can hold in memory. Knowing this matters for big tasks.

40 min · Reviewed 2026

The big idea

Each AI can only 'remember' a certain amount of text in one chat. This is the context window. Bigger window = can handle longer documents. Understanding this saves you from confusion.

Some examples

GPT-4 / Claude: huge context windows now (millions of tokens).
Smaller models: smaller windows.
If your chat gets very long, AI may forget early stuff.
For long documents, paste the whole thing at the START of the chat.

Try it!

Apply context window in your model-families workflow to get better results
Apply AI memory in your model-families workflow to get better results
Apply limits in your model-families workflow to get better results

Apply Context Windows: How Much AI Can 'Remember' in a live project this week
Write a short summary of what you'd do differently after learning this
Share one insight with a colleague

Why Gemini's 2M Context Window Matters (and Doesn't)

The big idea

Some examples

Pasting a 500-page PDF and asking 'what does it say?' — surface info works fine.
Asking 'find the one sentence about taxes on page 213' — performance drops noticeably.
For long docs, splitting into chunks and asking targeted questions often beats one big paste.
RAG (retrieval-augmented generation) tools like NotebookLM are designed exactly for this.

Try it!

Take a long article (10+ pages). Paste into Gemini and ask a specific question. Then split it into 3 parts and ask the same question. Compare answers.

Why Some Models Have 1 Million-Token Context Windows

The big idea

Some examples

You drop a 300-page novel into Gemini and ask 'what foreshadowing did chapter 3 contain about chapter 27?'
You paste a whole codebase into Claude and ask 'where would I add a feature flag?'
You paste 30 PDFs of research into Gemini and ask for a meta-analysis.
You paste a 2-hour transcript of a podcast and ask for chapter timestamps.

Try it!

Find a long document (a novel, a codebase, a course textbook). Paste as much as you can into Gemini or Claude. Ask a question that requires connecting two distant parts.

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-builders-model-families-AI-and-context-windows-teen

What is the core idea behind "Context Windows: How Much AI Can 'Remember'"?
1. Each AI has a 'context window' — how much it can hold in memory. Knowing this matters for big tasks.
2. Behave identically across families on the same prompt
3. Use for: cost-sensitive use cases at high token volume (Gemini Flash is competit…
4. Measure cache hit rates per provider
Which term best describes a foundational idea in "Context Windows: How Much AI Can 'Remember'"?
1. AI memory
2. context window
3. limits
4. Behave identically across families on the same prompt
A learner studying Context Windows: How Much AI Can 'Remember' would need to understand which concept?
1. context window
2. limits
3. AI memory
4. Behave identically across families on the same prompt
Which of these is directly relevant to Context Windows: How Much AI Can 'Remember'?
1. context window
2. AI memory
3. Behave identically across families on the same prompt
4. limits
Which of the following is a key point about Context Windows: How Much AI Can 'Remember'?
1. GPT-4 / Claude: huge context windows now (millions of tokens).
2. Smaller models: smaller windows.
3. If your chat gets very long, AI may forget early stuff.
4. For long documents, paste the whole thing at the START of the chat.
Which of these does NOT belong in a discussion of Context Windows: How Much AI Can 'Remember'?
1. Behave identically across families on the same prompt
2. Smaller models: smaller windows.
3. If your chat gets very long, AI may forget early stuff.
4. GPT-4 / Claude: huge context windows now (millions of tokens).
Which statement is accurate regarding Context Windows: How Much AI Can 'Remember'?
1. Apply AI memory in your model-families workflow to get better results
2. Apply limits in your model-families workflow to get better results
3. Apply context window in your model-families workflow to get better results
4. Behave identically across families on the same prompt
Which of these correctly reflects a principle in Context Windows: How Much AI Can 'Remember'?
1. Write a short summary of what you'd do differently after learning this
2. Share one insight with a colleague
3. Behave identically across families on the same prompt
4. Apply Context Windows: How Much AI Can 'Remember' in a live project this week
What is the key insight about "The rule" in the context of Context Windows: How Much AI Can 'Remember'?
1. Match the AI to the size of your task. Big task = pick AI with big context window.
2. Behave identically across families on the same prompt
3. Use for: cost-sensitive use cases at high token volume (Gemini Flash is competit…
4. Measure cache hit rates per provider
What is the recommended tip about "Match model to task" in the context of Context Windows: How Much AI Can 'Remember'?
1. Behave identically across families on the same prompt
2. Different models have different strengths. A coding model won't write great poetry; a creative model won't write tight S…
3. Use for: cost-sensitive use cases at high token volume (Gemini Flash is competit…
4. Measure cache hit rates per provider
Which statement accurately describes an aspect of Context Windows: How Much AI Can 'Remember'?
1. Behave identically across families on the same prompt
2. Use for: cost-sensitive use cases at high token volume (Gemini Flash is competit…
3. Each AI can only 'remember' a certain amount of text in one chat. This is the context window. Bigger window = can handle longer documents.
4. Measure cache hit rates per provider
What does working with Context Windows: How Much AI Can 'Remember' typically involve?
1. Behave identically across families on the same prompt
2. Use for: cost-sensitive use cases at high token volume (Gemini Flash is competit…
3. Measure cache hit rates per provider
4. Understanding "Context Windows: How Much AI Can 'Remember'" in practice: Understanding AI in this area gives you a real advantage in how you…
Which best describes the scope of "Context Windows: How Much AI Can 'Remember'"?
1. It focuses on Each AI has a 'context window' — how much it can hold in memory. Knowing this matters for big tasks.
2. It is unrelated to model-families workflows
3. It applies only to the opposite beginner tier
4. It was deprecated in 2024 and no longer relevant
Which section heading best belongs in a lesson about Context Windows: How Much AI Can 'Remember'?
1. Behave identically across families on the same prompt
2. Some examples
3. Use for: cost-sensitive use cases at high token volume (Gemini Flash is competit…
4. Measure cache hit rates per provider
Which section heading best belongs in a lesson about Context Windows: How Much AI Can 'Remember'?
1. Behave identically across families on the same prompt
2. Use for: cost-sensitive use cases at high token volume (Gemini Flash is competit…
3. Try it!
4. Measure cache hit rates per provider

← Back to interactive lesson