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 input with text responses
Manual extended thinking controlled by budget_tokens
Separate low, medium, and high effort settings
Manual interleaved thinking with the documented beta header
A 200K-token context, prompt caching, and batch processing
Before you choose
Opus 4.5 is a legacy model; it is not described here as the most intelligent or newest option.
Adaptive thinking, xhigh, and max effort are unsupported.
The synchronous output ceiling is 64,000 tokens, not the 128K or extended Batch ceilings of newer models.
Thinking can increase response time and billable output. It does not remove the need for source checks or code tests.
Extended thinking
lowmediumhigh · default
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. Opus 4.5 additionally supports low, medium, and high effort, defaulting to high. Effort shapes the overall response while budget_tokens controls manual thinking depth; set both when tuning a thinking request. Neither max nor xhigh is supported. Do not confuse the high effort default with thinking being enabled.
claude-opus-4-5 is a convenience alias for claude-opus-4-5-20251101.
Anthropic lists retirement not sooner than November 24, 2026. This is not a scheduled retirement or a deprecation announcement.
Manual interleaved thinking requires interleaved-thinking-2025-05-14 on supported integrations. Preserve returned thinking blocks as documented when continuing a tool conversation.
The reliable knowledge cutoff is May 2025, distinct from the August 2025 training-data cutoff.
The model has a 200,000-token context and 64,000-token output maximum; do not reuse the larger limits of Opus 4.6 or later.
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
Review a technical proposal
Request a structured critique grounded in constraints and source material.
Review this technical proposal against the stated goals and constraints. Identify unsupported assumptions, overlooked failure modes, and sections that need evidence. Cite the proposal paragraph behind each concern. Rank the three most consequential revisions and provide a test or measurement for each. Preserve the proposal's intent and do not invent requirements or recommend unrelated infrastructure.
Workflow 02
Edit a source-grounded narrative
Improve writing while preserving factual claims and the author’s voice.
Edit this draft for clarity, structure, and consistency while preserving the author's voice. Check every factual assertion against the attached source notes. Flag claims without support rather than strengthening them. Return a revised draft followed by a concise change log that separates style edits from factual corrections. Do not add quotes, statistics, or endorsements.
Workflow 03
Plan a safe component refactor
Keep a refactor limited to documented behavior and testable boundaries.
Plan a refactor of this component using its code, current tests, and public interface below. Identify duplicated logic and state transitions, then suggest the smallest sequence of changes that preserves behavior. Define a verification step after each change and flag unclear requirements. Include accessibility and failure states. Do not implement the plan or change external contracts.
Developer reference
Anthropic API pricing
These are Anthropic API reference prices, not EZ Ai Assist subscription prices.
The table covers the documented 200K context. Do not apply later models’ 1M-context or Fast-mode terms to Opus 4.5.
Regional partner deployments may have separate premiums and availability. The newer inference_geo API parameter is not supported on this model.
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.
Does Opus 4.5 use adaptive thinking?
No. It uses manual extended thinking with budget_tokens. Unlike Sonnet 4.5 and Haiku 4.5, it also supports a separate effort parameter. Adaptive mode returns an error on this generation.
What is the difference between effort and the thinking budget?
The budget controls manual reasoning depth. Effort shapes how much work the model puts into the overall response. On Opus 4.5 the supported effort levels are low, medium, and high, with high as the default.
Is thinking on at the default high effort?
No. Thinking is off until explicitly enabled. Set thinking.type: enabled and budget_tokens in a supported integration, then budget max_tokens for reasoning plus the final answer.
Is November 24, 2026 a shutdown date?
No. Anthropic’s documentation says retirement will be no sooner than that date. It is a commitment, not a scheduled shutdown or deprecation notice. Check the lifecycle source again before making migration decisions.
How does its context differ from Opus 4.6?
Opus 4.5 has a documented 200,000-token context and 64,000-token output ceiling. Opus 4.6 has larger limits and different thinking behavior. Do not copy limits or request settings across generations without verification.
Can I use it for publication-ready writing?
It can help revise and structure text, but review factual claims, citations, permissions, and tone yourself. The editorial examples on this page are starting points, not evidence that any generated draft is ready to publish.
Do these API rates determine my plan price?
No. They describe direct Anthropic API token categories. EZ Ai Assist subscriptions are separate, and available model controls may differ. Consult our pricing page and the app for current plan and access details.
Check the source
Official documentation
Specifications and API prices checked on . Example prompts and workflow advice are editorial guidance from EZ Ai Assist.