Every current Claude model ships a 1 million token context window with 128,000 tokens of max output: Claude Fable 5.2 (November 9, 2026), Claude Opus 5.5 (September 22, 2026), Claude Sonnet 5.5 (September 28, 2026) and Claude Haiku 5.5 (October 7, 2026). The older Claude Haiku 4.5, now a legacy model, runs 200,000 tokens with 64,000 tokens of max output, and Anthropic is removing it from the claude.ai model picker on Free, Pro and Max as of October 29, 2026. On the tokenizer Anthropic introduced with Claude Opus 4.7, one million tokens is roughly 555,000 words, enough to hold an 1,100-page book, an entire mid-sized codebase, or a year of chat history in a single prompt. Fable 5.2, Opus 5.5 and Sonnet 5.5 include the full 1M window at standard API pricing with no long-context surcharge, while Haiku 5.5 charges a higher rate for prompts over 100,000 tokens.
This guide explains what the Claude context window means, how many tokens each current model accepts, what "max output" caps your responses at, and what practically fits inside 1M tokens. Figures are verified against Anthropic's documentation and cross-checked with our token counter tool. The context window is the single most important spec for long-document and agentic work, so it is worth understanding before you paste a repository or a book into Claude.
What a Context Window Is and What It Does Not Do
A context window is the maximum number of tokens a model can consider at once, counting both your input and the model's output. When a conversation exceeds that limit, the oldest tokens fall out of the window and Claude effectively forgets them. A 1 million token window means Claude can hold about 555,000 words of combined prompt and response in working memory at one time. ChatGPT measures its ceiling differently: see the ChatGPT character and word limit guide.
What a context window does not do is give the model permanent memory. It resets between conversations, and once tokens scroll out they are gone unless you saved them. That is why searching and exporting past conversations matters: Claude's window is large, but it is still a window. Per Anthropic's documentation, tokens include words, sub-words, punctuation, and whitespace, so token counts run higher than word counts: roughly 0.555 words per token on Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 5.5 and the other models from Claude Opus 4.7 onward, which is about 1.8 tokens for every word.
Claude Token Limits Comparison Table (2026)
The table below lists the context window, max output, and API pricing for every current Claude model, verified for 2026. Pricing is per 1 million tokens via the Anthropic API and is separate from consumer Claude subscriptions.
Model
Context Window
Max Output
API Price (Input / Output per 1M)
Best For
Claude Fable 5.2
1,000,000 tokens
128,000 tokens
$8.00 / $40.00
Most capable generally available Claude model
Claude Opus 5.5
1,000,000 tokens
128,000 tokens
$4.00 / $20.00
Agents, hardest coding, deep reasoning
Claude Sonnet 5.5
1,000,000 tokens
128,000 tokens
$2.00 / $10.00
Balanced daily coding and writing
Claude Haiku 5.5
1,000,000 tokens
128,000 tokens
$0.10 / $0.50 (prompts up to 100K); $0.50 / $2.50 above
Fast, cheap, high-volume tasks
Claude Fable 5.1 (legacy)
1,000,000 tokens
128,000 tokens
$10.00 / $50.00
Still available; superseded by Fable 5.2
Claude Opus 5 (legacy)
1,000,000 tokens
128,000 tokens
$5.00 / $25.00
Still available; superseded by Opus 5.5
Claude Sonnet 5 (legacy)
1,000,000 tokens
128,000 tokens
$2.00 / $10.00
Still available; superseded by Sonnet 5.5
Claude Haiku 4.5 (legacy)
200,000 tokens
64,000 tokens
$1.00 / $5.00
Still available; superseded by Haiku 5.5
Two things stand out. All four current models, Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5, share the same 1M context window and the same 128K max output, so the choice between them is about reasoning quality, speed and cost, not input capacity or output length. Haiku 5.5 is the cheapest at $0.10 input / $0.50 output per 1M tokens on prompts up to 100,000 tokens, and the only one priced by prompt length. The legacy Haiku 4.5's 200K window is smaller but still holds roughly 150,000 words. Claude Sonnet 4.5, not listed above, is deprecated and retires on November 30, 2026, with Sonnet 5.5 as its replacement.
Context window in the Claude apps
In claude.ai chats on paid plans, Anthropic's help center lists a 1M context window for Fable 5.1, Opus 5.5, Opus 5, Sonnet 5.5, Sonnet 5 and Haiku 5.5; 500K for Fable 5, Opus 4.8, Opus 4.7, Opus 4.6 and Sonnet 4.6; and 200K for every other model. Anthropic's pricing page describes the context window on every plan as "Up to 1M, varies by model", so the window you get depends on the model you pick, not only on your plan.
What Fits Inside 1 Million Tokens
One million tokens is approximately 555,000 words on Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, which still covers almost any single document you would realistically paste into a chat. As a rule of thumb on those models, 1,000 words costs about 1,800 tokens, so the arithmetic is roughly words times 1.8.
That ratio surprises people who have used Claude for a while, and it is worth knowing why. Anthropic shipped a new tokenizer with Claude Opus 4.7, and its own pricing documentation states the change plainly: the newer tokenizer "produces approximately 30% more tokens for the same text." Every model from Claude 4.7 onward uses it, which covers Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 5.5, Opus 5 and Sonnet 5. The legacy Claude Haiku 4.5 shipped before the change and still uses the previous tokenizer, so its 200,000 token window holds roughly 150,000 words, at about 1,333 tokens per 1,000 words. Anthropic's own figure for the current tokenizer is 555,000 words, or about 2.5 million Unicode characters, per 1 million tokens.
Concretely, 1M tokens holds any one of these: an 1,100-page technical manual, an entire medium-sized software repository, a full novel plus its outline and worldbuilding notes, or dozens of research papers at once. That is why long-context work, such as reasoning across a whole codebase or synthesizing a large research corpus, is practical on Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5. The practical constraint is usually cost and latency, not the window size.
Running long Claude sessions and losing track of what you asked last week? AI Toolbox (formerly Claude Toolbox) adds full-text search, message bookmarks, and a context window meter to Claude so nothing scrolls out of reach. Trusted by 40,000+ users with a 4.7/5 Chrome Web Store rating. Install AI Toolbox free ->
Max Output vs Context Window: The Difference
The context window is the total token budget for input plus output; the max output is a separate, smaller cap on how long a single response can be. Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 accept 1M tokens of input and generate up to 128,000 tokens in one reply. The legacy Haiku 4.5 caps at 64,000 output tokens.
This distinction matters for generation-heavy tasks. If you ask Claude to produce a very long document, a full codebase, or an exhaustive report, the max output ceiling, not the context window, limits the reply length. Every current model generates up to 128K tokens, which is ample for long structured generation; only the legacy Haiku 4.5 caps lower at 64K. For inputs, the 1M window is almost never the bottleneck for a single document.
1M Context at Standard Pricing: Why It Matters
Opus 5.5 and Sonnet 5.5 include the full 1 million token context window at standard API pricing, with no long-context surcharge, as does Fable 5.2. Haiku 5.5 is the exception: prompts over 100,000 tokens pay $0.50 / $2.50 per 1M instead of $0.10 / $0.50. Previous generations, including Opus 4.8 and Sonnet 4.6, are now legacy, and older models capped near 200K with extra charges for the top of the window. The current lineup removes both constraints.
What this changes: you stop engineering around context limits, with fewer retrieval pipelines and chunking strategies, and you can ask questions that span an entire corpus in a single pass. At $2 input on Sonnet 5.5, a full 1M-context prompt costs about $2 in input tokens plus output, and about $0.50 on Haiku 5.5, which is inexpensive for the capability.
Why Claude Forgets and How to Keep Your History
Claude forgets earlier parts of a long conversation when the total tokens exceed the context window, and it forgets everything between separate conversations because the window resets. The model has no persistent memory of past chats beyond what you paste back in or store in a Project.
To keep important conversations reachable, you need to search and export them. Claude's native tools are limited: a chat search with no date, role or exact-match filters on all plans, conversational RAG search on paid plans, and export only as a full-account dump with a 24-hour link. AI Toolbox adds full-text search across every synced message with an exact-match toggle, message-level bookmarks with scroll-to and highlight, per-conversation TXT and JSON export, and a live context window meter. It works on the free Claude plan. See ai-toolbox.co/claude-toolbox.
Limitations and What Native Claude Handles Well
Claude's 1M context window is genuinely generous, and for most single-document tasks you will never hit it. Native Claude handles pasting a long document, reasoning over a large codebase, or holding a Project's knowledge base without any extension. If your workflow is one long input at a time, native Claude is enough.
The limitation is across conversations, not within them. Once a chat ends or the window fills, native Claude gives you no fast way to find an exact phrase or export a single clean conversation. Token counts are also estimates that vary with language and formatting, so treat the 1.8 tokens-per-word ratio as a guide. For exact counts, use Anthropic's tokenizer.
Frequently Asked Questions
What is Claude's context window in 2026?
Every current Claude model, Claude Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5, has a 1 million token context window, while the legacy Claude Haiku 4.5 has 200,000 tokens. One million tokens is roughly 555,000 words of combined input and output on the tokenizer these models use, large enough to hold an 1,100-page book, an entire codebase, or several novels in a single prompt. Fable 5.2, Opus 5.5 and Sonnet 5.5 include 1M at standard pricing; Haiku 5.5 charges more for prompts over 100,000 tokens.
How many words is 1 million tokens for Claude?
Approximately 555,000 words on Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, which is the figure Anthropic publishes for the tokenizer it introduced with Claude Opus 4.7. Note the direction: a word costs about 1.8 tokens on those models, so 1,000 words is around 1,800 tokens, not 555. The legacy Claude Haiku 4.5 predates the change and runs the older ratio, where 1,000 words is about 1,333 tokens and its 200,000 token window holds roughly 150,000 words. The exact count varies with word length, punctuation, and special characters, so use a token counter for precise figures before pasting large inputs.
What is the maximum output length for Claude?
Claude Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5 generate up to 128,000 tokens per response, while the legacy Haiku 4.5 caps at 64,000 tokens. Max output is separate from the context window and limits how long a single reply can be, not how much you can input. For very long outputs, use any current model, or request the content in sections.
Do all Claude models have the same context window?
Not all of them. The current models, Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5, share a 1 million token context window, but the legacy Haiku 4.5 runs 200,000 tokens. The current models share the same 128K max output but differ on price and speed. In claude.ai chats on paid plans, Fable 5, Opus 4.8, Opus 4.7, Opus 4.6 and Sonnet 4.6 get 500K, and other older models get 200K.
Why does Claude forget earlier messages?
Claude forgets earlier messages when a conversation exceeds its context window; the oldest tokens drop out of working memory. It also resets between separate conversations. To preserve important chats, search and export them with AI Toolbox before they scroll out of the window, or store key material in a Claude Project.
How do I reduce token usage in Claude?
Be concise in prompts, avoid repeating context, and break long tasks into separate conversations. Paste only the sections you need rather than entire documents when possible, and use a cheaper model like Haiku 5.5 for high-volume steps. AI Toolbox's Prompt Library lets you save optimized, reusable prompts so you spend fewer tokens re-explaining context.
Bottom Line
In 2026, every current Claude model, Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5, ships a 1 million token context window (roughly 555,000 words on Anthropic's current tokenizer) and 128,000 tokens of max output, while the legacy Haiku 4.5 runs 200K, about 150,000 words, with 64,000 tokens of output. The window is large, but it is still a window: it resets between chats and drops old tokens when full. AI Toolbox adds search, bookmarks, export, and a context window meter so your Claude history stays reachable, free to start, $9.99/month, or $99 lifetime. From the makers of AI Toolbox (formerly ChatGPT Toolbox) (40,000+ users, 4.7/5 rating).
A Full Stack Developer with 7+ years of experience building AI productivity tools. Leads product development and frontend architecture for AI Toolbox, the Chrome extension suite (ChatGPT, Gemini, Grok, and Claude modules) that helps users search, organize, and export their AI conversations.
Liked this guide? Get the next one first.
Join 40,000+ readers getting practical prompts, new features, and subscriber-only deals. No spam, unsubscribe in one click.
New features and product updates before anyone else
Time-saving prompts and workflows for ChatGPT, Gemini, Claude, and Grok