Back to models
LegacyChat Models

Anthropic

Claude Opus 4.7

Best for Research

Anthropic’s legacy Opus model with adaptive thinking, five effort levels, and a 1M-token context for supplied material.

1M contextAgentic
Open AppBrowse All Models

At a glance

Know the model before you prompt.

Anthropic API specifications
Context window
1,000,000 tokens
Maximum output
128,000 tokens
Inputs → output
Text + Images → Text
Knowledge cutoff
January 2026

API model ID: claude-opus-4-7

Capabilities & boundaries

What it supports. Where the limits are.

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 understanding with text responses
  • Adaptive thinking with low through max effort
  • Automatic thinking between integrated tool calls in adaptive mode
  • A 1M-token context with 128,000-token synchronous output
  • Prompt caching and asynchronous Message Batches

Before you choose

  • Opus 4.7 is legacy, not the newest Opus model.
  • Manual budget_tokens thinking is unsupported, and non-default sampling parameters are rejected.
  • Fast mode is not supported; requesting speed: fast returns an error rather than a faster response.
  • Model API limits do not guarantee equivalent tools, inputs, or quotas in EZ Ai Assist.

Adaptive thinking and effort

lowmediumhigh · defaultxhighmax

Thinking is off by default. Enable thinking.type: adaptive and choose effort separately; high is the effort default. Anthropic recommends xhigh as a starting point for coding and agentic tasks. Opus 4.7 follows low and medium effort more strictly than Opus 4.6, so evaluate shallow answers before reducing effort. Reserve max_tokens for both thinking and the final answer. Manual enabled mode and between_tools are unsupported.

  • The synchronous output limit is 128,000 tokens; 300,000-token output requires the Message Batches beta with output-300k-2026-03-24.
  • The documented retirement commitment is not sooner than April 16, 2027, not a scheduled retirement.
  • The new tokenizer can produce about 30% more tokens from the same text than pre-4.7 models. Recheck budgets, caching, and context utilization.
  • Setting speed: fast returns an error on Opus 4.7. Do not copy Fast-mode settings from Opus 4.8 or later models.
  • Remove non-default temperature, top_p, and top_k parameters when migrating; use clear instructions and measured effort settings instead.

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

Check an API migration diff

Find compatibility risks without conflating model settings across generations.

Compare this old API request and proposed migration diff against the documentation excerpts I provide. List each changed parameter, its purpose, and any incompatibility. Flag settings with no supporting documentation. Propose a minimal valid request and a test matrix for response shape, stop conditions, token usage, and tools. Do not send the request or insert credentials.

Workflow 02

Design a contained agent task

Define a bounded coding task with explicit permissions and measurable stopping conditions.

Turn this maintenance request into a bounded agent task. Identify the relevant files from the supplied index, define acceptance tests, and list actions that require approval. Set clear stopping conditions for missing evidence, unexpected changes, or tool failure. Provide a review checklist and rollback approach. Do not execute the task or expand it beyond the requested maintenance.

Workflow 03

Compare document revisions

Request precise citations for differences that could affect implementation.

Compare these two revisions of a technical specification. Group differences into behavior, data format, dependencies, and unresolved decisions. Cite the section in each revision for every change and explain its likely impact without inventing downstream systems. Highlight contradictions and propose clarifying questions. End with a compact implementation checklist tied to the approved revision.

Developer reference

Anthropic API pricing

These are Anthropic API reference prices, not EZ Ai Assist subscription prices.

View EZ Ai Assist plans
Standard processing · USD per 1,000,000 tokens
Token typePrice
Input$5.00
5-minute cache write$6.25
1-hour cache write$10.00
Cache read$0.50
Output$25.00
  • Standard prices cover the full 1M context window with no long-context surcharge.
  • No Fast mode is available for Opus 4.7. Do not apply another model’s Fast-rate table.
  • US-only inference_geo processing has a 10% token-rate premium where supported; partner regional rates are separate.
  • 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.

What distinguishes Opus 4.7 from earlier Claude 4 models?

Its documented changes include adaptive-only thinking when enabled, a newer tokenizer, and stricter request-parameter rules. It also supports xhigh effort. These differences require testing; they are not a reason to assume every old request can be replayed unchanged.

Is Opus 4.7 the current flagship?

No. Its overview marks it legacy and points to Opus 5.5 as a migration option. The legacy label is distinct from deprecation or retirement, and no shutdown date is assigned in the reviewed lifecycle table.

Which effort should I start with?

The API default is high. Anthropic recommends starting at xhigh for coding and agentic work, while measuring quality and token use. Lower effort can be appropriate after evaluation, and thinking must still be explicitly enabled.

Can I keep budget_tokens from an older integration?

No. Opus 4.7 rejects manual extended thinking. Replace the old configuration with adaptive thinking and supported effort controls, then re-evaluate rather than treating effort as an exact token-budget conversion.

Does Opus 4.7 support Fast mode?

No. The reviewed documentation says speed: fast returns an error. Fast support on a neighboring Opus model does not make it available here, and the app may expose a different set of controls.

How should I estimate long-document cost?

Count tokens with the actual model tokenizer and distinguish uncached input, cache writes, cache reads, and output including thinking. The full context uses standard rates, but a new tokenizer can change the number of billable tokens.

Can these prompts operate tools or change my files?

The example text alone does not grant tools or permissions. Each prompt is a reviewable starting point and asks for analysis or a plan. Any real tool execution depends on the integration and the authorization you provide.

Check the source

Official documentation

Specifications and API prices checked on . Example prompts and workflow advice are editorial guidance from EZ Ai Assist.

Same provider

More from Anthropic

Anthropic

Claude Sonnet 5.5

Anthropic model for coding and everyday professional work, with adaptive thinking, image input, and a 1M-token context window.

Anthropic

Claude Opus 5.5

Claude Opus 5.5 is Anthropic’s model for demanding coding and analysis, with always-on adaptive thinking and optional API Fast Mode.

Anthropic

Claude Opus 5

Anthropic’s active legacy Opus model for complex coding and analysis, with adaptive thinking and a 1M-token context window.

Anthropic

Claude Sonnet 5

Anthropic’s active legacy Sonnet model for coding and general work, with adaptive thinking and a 1M-token context window.