Tool support requires the appropriate API integration; a supported tool is not automatically active in every chat.
Supported API features and tools
Text and image inputs with text output
Manual extended thinking with a configurable token budget
A 200,000-token context and 64,000-token output ceiling
Interleaved manual thinking through the documented beta configuration
Prompt caching and discounted Message Batches
Before you choose
Deprecated and scheduled to retire on Anthropic-operated platforms; partner schedules can differ.
Adaptive thinking and the effort parameter are not supported on this model.
Do not inherit the 1M context or extended Batch output of newer Sonnet models. This guide uses the current overview’s 200K standard context.
Native output is text. Tools require integration, and model access in EZ Ai Assist must be checked separately.
Extended thinking
Thinking is off by default. On supported integrations, enable manual extended thinking with thinking.type: enabled and budget_tokens. Start with a modest budget and evaluate the result; max_tokens must also leave room for the final answer. The ordinary budget must be at least 1,024 tokens and below max_tokens; interleaved tool thinking has separate budget rules. Adaptive thinking is not supported. The effort parameter is not supported. For manual reasoning between tool calls, the documented interleaved-thinking beta header is required. When migrating to Sonnet 5.5, evaluate its different thinking and effort behavior rather than carrying over a fixed budget unchanged.
claude-sonnet-4-5 is an alias for claude-sonnet-4-5-20250929. Use the exact snapshot when auditing older requests.
Anthropic’s dates apply to its operated platforms; partner-operated Amazon Bedrock and Google Cloud maintain separate retirement schedules.
The reliable knowledge cutoff is January 2025; the training-data cutoff is July 2025.
The reviewed model overview lists a standard 200,000-token context and 64,000-token output limit. Any historical long-context beta must be verified separately before relying on it.
The interleaved-thinking-2025-05-14 beta header enables manual interleaving on supported integrations; header acceptance and behavior can differ by platform.
Put it to work
Start with a more useful prompt.
Original examples from EZ Ai Assist. Adapt these to your task and the features available in your workspace.
Workflow 01
Inventory a Sonnet migration
Make model usage and request settings visible before a planned retirement.
Review this anonymized request inventory for Sonnet 4.5. Group uses by prompt type, tool dependencies, output format, and risk. Identify request settings that need checking against the supplied Sonnet 5.5 documentation. Produce a staged migration checklist with owners, test cases, and rollback conditions. Do not change traffic or infer that a successful sample proves complete compatibility.
Workflow 02
Create behavior-preserving tests
Turn a bug fix into a focused regression suite that works across model comparisons.
Using this bug report, minimal reproduction, and patch, propose regression tests that distinguish the old behavior from the intended fix. Include boundary values, failure paths, and an assertion tied to each requirement. Cite the relevant code when explaining a risk. Separate tests that can run locally from integration checks. Do not rewrite the patch or claim execution results.
Workflow 03
Review a support escalation
Use source-grounded drafting without inventing customer promises.
Draft a support escalation from the customer report and approved troubleshooting notes below. Summarize the observed issue, attempted steps, remaining uncertainty, and the information engineering needs next. Cite the notes supporting each recommendation. Keep the tone calm and avoid inventing deadlines, refunds, or capabilities. Remove unnecessary personal data and flag any action needing approval.
Developer reference
Anthropic API pricing
These are Anthropic API reference prices, not EZ Ai Assist subscription prices.
These are the published standard reference rates while the model is deprecated, not a promise of availability after November 30, 2026.
The table covers standard 200K-context usage. Historical long-context beta conditions and partner pricing must be checked separately.
Thinking tokens are billed as output, even when only a summary or no thinking text is displayed. Budget for the complete output usage, not just the visible answer.
Batch processing discounts input and output by 50%. Cache writes and reads have separate rates and eligibility; tools can add fees.
Prices and platform availability can change. Confirm the current provider, region, processing tier, and cache behavior before estimating direct API spend.
Common questions
A few things worth knowing.
When does Sonnet 4.5 retire?
Anthropic announced deprecation September 30, 2026, with retirement November 30, 2026 on its operated platforms. Partner platforms set their own dates. Verify your actual provider and app access rather than assuming all services shut down together.
Which model should replace it?
Anthropic recommends Claude Sonnet 5.5. Test your actual prompts, tools, structured response handling, and latency requirements before moving traffic. A replacement name alone does not prove equivalent behavior.
Can I enable adaptive thinking on Sonnet 4.5?
No. This model supports manual extended thinking with budget_tokens, not adaptive mode. The effort parameter is unsupported too. Those controls change when migrating to a newer Sonnet model.
Does it have a 1M context window?
The current reviewed overview lists a standard 200,000-token context and 64,000-token maximum output. Older beta arrangements are not assumed here. Check the exact provider and API feature terms if your integration used a larger context.
What happens to existing links after retirement?
The marketing guide remains available as a historical and migration reference. That does not keep a retired API model operational. This page’s lifecycle notice will need a source-backed update when the retirement takes effect.
How should I verify the migration?
Create a representative evaluation set including failure cases, permissions, citations, and output parsing. Compare acceptance rates, token use, and response time. Keep a staged rollout and a recovery plan rather than relying on a few successful examples.
Are the listed rates part of my subscription?
No. Direct API reference prices are separate from EZ Ai Assist plans. Use the pricing page and the app for subscription details, and treat the sample prompts as editorial starting points requiring human review.
Check the source
Official documentation
Specifications and API prices checked on . Example prompts and workflow advice are editorial guidance from EZ Ai Assist.