What is a context window, and what is a 1 million token context good for?
In short: The context window is the upper limit on what a model can see in a single request: the system instruction, earlier messages in the chat, the files you attach and the answer the model will write all have to fit. As of October 30, 2026, on the Wiozen API wiozen-anthropic has a 200K window, and the Claude 5.5 models, Gemini 3.8 Flash and DeepSeek V4.1 Flash have 1M tokens. A big window doesn't mean you should send everything; every token you send is charged.
What goes into the window?
The model doesn't remember earlier conversation. Even when a chat app or coding agent looks like it's continuing a conversation, the tool sends the entire history again with every new message. Whatever is sent in that request goes into the window: the system instruction, earlier questions and answers, tool definitions and tool output, attached files and images.
The model's answer takes its share of the same window. With a 200K window and 190K tokens of input, only 10K is left for the answer. So think of the limit as "input plus requested output".
How big are the windows of the models on the Wiozen API?
| Model | Context window |
|---|---|
wiozen-anthropic | 200K |
claude-opus-5-5, claude-sonnet-5-5, claude-haiku-5-5 | 1M |
gemini-3.8-flash-high | 1M |
deepseek-v4.1-flash | 1M |
The table is from the October 30, 2026 catalog; the current value is in the Context column on the Models page. Output has its own limit: a single answer can be at most 262,144 tokens (256K), or the model's own output limit if that's lower. The request body can be at most 32 MB; base64-embedded images and long files are what get close to that fastest.
What is a 1 million token window good for in practice?
A big window matters for work that has to be read in one piece: reviewing a mid-sized code repository in one go, looking for contradictions in a contract hundreds of pages long, summarizing a long support log. Splitting a job that doesn't fit into 200K and asking about each part separately is tedious, and it misses the links between the parts.
For everyday chat, a small code fix or a one-page text, a big window adds nothing. The input there is a few thousand tokens anyway, and price and quality decide which model to use, not window size.
What happens when the window is full?
If the input you send plus the output you ask for exceeds the window, the model rejects the request and returns an error. Failed requests aren't charged. The Wiozen API doesn't trim old messages for you; the tool you use, or your own code, decides what gets sent.
Most coding agents summarize the conversation or drop old tool output themselves when the window gets close to full. If you do it in your own code, remove the oldest messages and file contents you no longer need, and keep the system instruction and the last few messages. To see how many tokens you send, look at the usage field in the response or use the free /v1/messages/count_tokens endpoint in the Anthropic format.
Why is long context expensive?
Every token that goes into the window in a request is charged as input. Ask ten questions about a 300K-token document and the document is sent ten times and charged ten times. As a conversation grows, each new message costs more than the one before, because the history goes out again every time.
For one model the effect is sharper: on claude-haiku-5-5, if a request's total input (cache reads included) goes over 100K tokens, every item in that request is priced at five times the rate. At the October 30, 2026 catalog price, input goes from $0.07 to $0.35 per 1M tokens and output from $0.35 to $1.75. If you plan to use Haiku with long documents, keeping input under 100K or picking a model whose price doesn't change with long input is cheaper.
If you send the same long input again and again, caching cuts the cost a lot: on most models cached input costs a tenth of normal input or less. We walk through the numbers in how to calculate AI API cost.
Why might my balance not cover a large input?
When a request starts, its estimated cost is held from your balance. The estimate comes from the size of the body you send and from max_tokens, and since nobody knows in advance how much will be read from cache, all input is counted at the normal price. A 1M-token request on an expensive model means an estimate of a few dollars. If your balance doesn't cover it, the request doesn't start and you get 429 Quota exhausted; once an answer finishes, the real amount is written and the difference goes back to your balance.
Frequently asked questions
Does the model remember my earlier conversation?
No. Your tool sends the history again with every request. The model doesn't see messages that aren't sent.
Does the answer count toward the context window?
Yes. Input plus requested output has to fit. On top of that, a single answer can be at most 256K tokens.
Which model has the largest context window?
As of October 30, 2026, the Claude 5.5 models, Gemini 3.8 Flash and DeepSeek V4.1 Flash, at 1M tokens. wiozen-anthropic has 200K.
Is a request that exceeds the window charged?
No. The model rejects it, and failed requests aren't charged.
Does the price go up with long context?
On most models the per-token price stays the same; you just pay for more tokens. The exception is claude-haiku-5-5: once total input goes over 100K tokens, that request is priced at five times the rate.
Compare window sizes and prices on the Models page.
More guides: How can you use an AI API from Turkey without a credit card? · How do you connect an OpenAI-compatible API to a tool you already use? · How do you calculate the cost of an AI API? · What should I do if my AI API key has leaked? · How do you generate images with an AI API, and what does one image cost? · Why does my AI API return a 429 error, and how do I fix it? · Why do long AI API requests time out: streaming or non-streaming? · Anthropic Messages API or OpenAI Chat Completions: which one should I use? · Which AI model should I choose: deciding by price, context and quality · Why does a reasoning model return an empty answer, and why are thinking tokens charged? · Why is my API balance running out faster than expected? Tracking spend request by request · How do you convert a web page to Markdown to feed it to an AI model? · How do you get AI to write and publish articles in WordPress automatically? · How do you find blog topics from RSS feeds and Google Trends? · Does AI-written content show up on Google and in AI answers? · How do you get internal links into AI-written WordPress articles? · How do you get AI to generate a different featured image for every WordPress post? · How do you write a content brief when an AI writes your blog posts? · What should you watch for when an AI writes recipes for your food blog? · How do you fix errors when connecting a WordPress plugin to your AI API account? · Why does my AI article bot stop halfway in WordPress? "Run is stuck" and LiteSpeed · Why are my scheduled WordPress posts late? WP-Cron and a real server cron · How do I get an AI article bot to write blog posts in English or another language? · What does "Content Bot cannot generate articles" mean? Reading the bot's log