Chat with Claude Sonnet 4.6

AI ChatClaude Sonnet 4.6

What is Claude Sonnet 4.6?

Claude Sonnet 4.6 is Anthropic's model for coding, document analysis and tool-assisted work, released February 17, 2026. Its API ID is claude-sonnet-4-6. It accepts text and images and returns text. As of September 8, 2026, the latest Sonnet is Claude Sonnet 5, released June 30. The product page retains a 4.6 heading while its availability section describes 5. This page stays with the still-available 4.6 so you can evaluate an existing workflow or a migration.

Sonnet 4.6 API facts today

Long input, bounded output: The model overview lists 1M context and 128K maximum output tokens for ordinary API requests. A separate Batch API beta allows more output. Label documents and reserve response space; fitting a file into context does not guarantee every detail will be retrieved.
Adjust reasoning to the task: Sonnet 4.6 supports adaptive thinking and defaults to high effort, but omitting the thinking field runs without thinking, as the migration guide clarifies. Manual extended thinking is deprecated. For small patches, test lower effort where available; measure time, accepted changes and billed usage together.
Price a complete attempt: Standard API rates are USD $3 per million input tokens and $15 per million output tokens. A request with 20,000 input and 4,000 billed output tokens costs $0.12 before caching, tools or other adjustments. These are provider rates, not Ottermind prices; repeated attempts increase the bill.

Four practical tasks for Sonnet 4.6

Start with a bounded piece of work and an observable finish line. These tasks give you something to run, compare or inspect instead of relying on the model's statement that it is done.

Trace a bug across files
Provide the error, reproduction steps and relevant callers. Ask for the first incorrect assumption, a small patch and a test that fails before the fix. Inspect whether it changed the expected behavior or merely adjusted the test to accept a bug.
Implement one feature from a brief
Supply the interface contract, sample data and acceptance cases. Ask for one working slice with real data boundaries, then review it before adding the next. Explicitly distinguish a prototype from production behavior so invented mocks do not pass as a finished integration.
Reconcile a document pack
Share dated reports or policy versions with stable page labels. Request a table of conflicting claims, supporting passages and unresolved gaps. Verify each cited passage and calculate important figures independently before turning the table into a decision brief.
Plan a multi-step form workflow
Provide the fields, decision rules and examples of missing information. Ask for a step-by-step plan with exception paths and a review point before submission. Actual browser actions require an application with compatible tools; a text plan does not mean a form has been submitted.

Claude Sonnet 4.6 vs Claude Sonnet 5

Decision point	Claude Sonnet 4.6	Claude Sonnet 5
Standard input / output	USD $3 / $15 per million tokens	USD $2 / $10 per million tokens
Context / ordinary max output	1M / 128K tokens	1M / 128K tokens
Current position	Available legacy model	Latest Sonnet model
Why evaluate it	Preserve a tested 4.6 workflow	Test improved execution at lower token rates

Choosing a version for your work

Where 4.6 remains useful

A known behavior baseline: Keep a set of reviewed 4.6 outputs when existing prompts or integrations depend on its behavior. Include difficult cases and failures in your evaluation. Familiarity can justify a staged migration, but it does not establish that the older model is better.
From findings to the next deliverable: Use Ottermind to carry a debugging brief, evidence table or workflow plan into the next stage. Keep source references and unresolved assumptions attached so revisions remain traceable. The models and tools you can use depend on your account.

Tradeoffs worth testing

Lower rates do not predict the whole bill: The release notes confirm that Sonnet 5's $2/$10 rates became permanent on August 10. Its new tokenizer produces approximately 30% more tokens for the same text, depending on the workload. Compare actual task cost, retries and latency rather than assuming a one-third saving.
Migration includes request changes: The Sonnet 5 overview says adaptive thinking is on by default. Manual extended thinking and non-default temperature, top_p or top_k return a 400 error. Check request settings and output quality before replacing the model ID in an existing integration.

What users actually reported

These individual reports are not a consensus or a controlled benchmark. They identify useful checks for your own workload; launch experiences and client-specific limits should not be treated as today's universal behavior.

More independent work, with assumptions to catch

In the Hacker News launch discussion, nichochar described longer independent app-building runs in Mocha's own agent setup. Another user, coder4rover, reported mock data and functions that needed a correcting prompt. Test whether longer runs finish the requested integration, not just a convincing demonstration.

API capacity and subscription access differ

A Reddit user on Claude Code v2.1.83 reported 200K context under a Max subscription and extra charges for 1M Sonnet use. This is a historical plan-specific report, not evidence that today's API is limited to 200K. Check the actual client, plan and usage settings.

Continue your task in Ottermind

Define scope and evidence

Include the files or passages that matter, the desired output and the checks for success. State which interfaces, facts or behaviors must remain unchanged.

Confirm the model in Studio

Continue to Studio and sign in if needed. Select Claude Sonnet 4.6 if available and review usage terms. This page does not automatically select the model or guarantee access.

Review before continuing

Run the reproduction, open cited passages or inspect a workflow's exception path. Bring accepted findings and remaining questions together into the next revision.

Claude Sonnet 4.6 FAQ

Is Sonnet 4.6 still available?

As of September 8, 2026, the API overview marks it Active (legacy), with retirement not sooner than February 17, 2027. That is not an announced shutdown date. Check the lifecycle table and your provider's availability before committing a workflow.

Does 1M context still require the launch beta?

No. The current context guide lists 1M as the default API window with standard pricing and no beta header. The February launch beta is historical. A subscription application can expose a different limit; confirm its terms separately.

Can Sonnet 4.6 create images or operate my browser?

It accepts text and images and outputs text. Browser actions and file changes need application tools. The migration guide lists computer_20251124 for both versions; 4.6 does not support Sonnet 5's newer computer_toolset_20260801 or browser-use tool. A written plan is not proof of completed actions.

Is chatting here free, and is the model selected?

This page promises neither free usage nor access to Sonnet 4.6. Your prompt continues to Studio; availability, tools and charges depend on your account. Confirm the selected model there before starting.

Bring a task you can check

Continue in Ottermind Studio with clear evidence, a bounded goal and room to revise the result.