Claude Sonnet Context Window: Check the Version and App
The usable Claude Sonnet context window depends on the exact model and where you use it. Check the right specifications, leave room for the answer, and preserve your project between sessions.
Updated 5 min read
The context window for Claude Sonnet 4.6 is the amount of material it can work with in one request, including room for its response. To find your usable limit, match the exact Sonnet version with the app or API route you are using; a model specification alone does not describe every Claude experience.
That distinction matters when a long document fits through one route but another session needs a shorter input. fenbs keeps the reviewed project record outside the conversation, so changing Sonnet versions does not require reconstructing every decision from chat history.
Which Sonnet version are you actually using?
Start with the identifier, not just the word Sonnet. Anthropic publishes separate specifications for Claude Sonnet 4.6 (opens in a new tab) and Claude Sonnet 5 (opens in a new tab). Open the page that matches the model selected in your app or sent in your request, then check both context capacity and maximum output.
A family name describes a group of models. It does not prove which member a particular tool selected. When asking a teammate to reproduce a failure, include the exact model identifier, the application, and whether the work involved a chat, uploaded document, or coding session.
| What you see | What to verify | Why it matters |
|---|---|---|
| Sonnet 4.6 | Match the request to claude-sonnet-4-6 and its model page. | An older tutorial may describe different access conditions. |
| Sonnet 5 | Match the request to claude-sonnet-5 and its model page. | Do not carry an earlier version's assumptions forward. |
| Sonnet without a version | Check the actual model selected by the app or service. | The label alone is insufficient for a capacity comparison. |
Keep the observation narrow: a request failed in this setup with these inputs. That gives you something testable. For broader differences in what the model families are useful for, see Sonnet versus Opus.
Why can the app and API show different limits?
The API, a developer connection to the model, and a consumer chat app are different access routes. Anthropic maintains a separate context-window guide for Claude apps (opens in a new tab). Use that guide for the app you are working in rather than treating the model page as an account entitlement.
In Claude Code, check the selected model and context settings against the model configuration documentation (opens in a new tab). If your organization uses another provider or a gateway, an intermediary that forwards model requests, also check that service's configuration. An application label is less useful than the settings governing the actual request.
For a repeatable check, send the same small task and source excerpt through the route you intend to use. Record the route beside the result. A successful test in a browser chat does not establish what a separate integration will accept, and a failed integration does not prove the underlying model has a smaller window.
What takes up space besides your prompt?
Tokens are the small pieces of text a model processes. Your latest question is only part of its workload: instructions, earlier messages, documents, tool information, and the generated response also matter. Anthropic's context-window documentation (opens in a new tab) explains how these components share the available space.
A file that fits at the start of a session may become harder to accommodate after many exchanges. Avoid estimating from page count alone. Dense tables, source code, and image-heavy documents are different kinds of input, and the answer you request needs space of its own.
- Define the current question before choosing which sources to include.
- Remove duplicate drafts and tool output that no longer helps the next action.
- State the required answer format so a short decision does not become an unnecessary report.
- Keep essential requirements in a concise, reviewed record rather than burying them in repeated conversation.
Our AI context window guide explains the working-memory idea in more detail. For project continuity, fenbs AI context gives a connected assistant the rules, current work, and Where we left off to read before it starts.
How should you compare versions for real work?
Use a task you can evaluate, not just a large file that uploads successfully. Say you are comparing revisions to a product specification. Ask each session to identify changed requirements, cite the relevant section, and list conflicts it cannot resolve. Check those answers against the originals.
Include a requirement near the beginning and another near the end, but keep the test representative of your actual work. If the answer omits something important, investigate source readability, instructions, and ambiguity before assuming that capacity is the only cause. A correct reference to one paragraph does not establish complete coverage of the whole document.
Keep neighboring questions separate: Claude Projects context covers project material, Claude context size covers the wider family, Claude versus Gemini compares providers, and the Opus context guide covers that model family.
What should survive when you change sessions?
Preserve the project's state rather than every sentence from the conversation. Save the current goal, the approved requirements, source locations, completed work, unresolved questions, and the next action. Label assumptions clearly so the next assistant does not mistake them for decisions.
Review this record before relying on it. If a summary says a design was approved, check who approved it and what version they saw. Ask the next assistant to explain the remaining work in its own words before it makes changes. The explanation is a useful check that the starting material is sufficient.
With fenbs connections, supported AI apps read the same saved work when you switch models. This preserves the record you maintain; it does not automatically transfer a Claude conversation or enlarge Sonnet's context window.
Keep the useful project context ready
Save the requirements that still apply, the open tasks, and the next step in fenbs. Ask your connected assistant to read that record before loading the source material for its current question. Keep a short starting point that remains useful when a chat ends or a different model takes over.
Create a free accountFree for one person · No credit card · Works with Claude, ChatGPT, Cursor and more
How to do this with fenbs
- Create a free account, then add the project whose work you need to carry between Sonnet sessions.
- Add the current tasks and record the decisions people approved in Decisions and rules. Include the source location needed to verify each requirement.
- Connect your supported AI app over MCP, a standard way for an assistant to use an outside tool, and ask it to read the project context before working.
- Before stopping, ask the assistant to save progress and Where we left off. Review the completed work, remaining questions, and next step before starting another session.
fenbs is free for one person, with no credit card. It is useful when work continues across conversations; if you only need to analyze a single document once, a focused Claude chat may be enough. fenbs does not host a model or import your Claude chat history.
After the first restart, check that the assistant can name the current requirement and next action without a long explanation from you. Correct stale records as the work changes. More guides in context engineering cover choosing the right material for each session.