Claude Opus 4.7

What is Claude Opus 4.7?

Opus 4.7 was Anthropic's flagship Opus model until Opus 4.8 replaced it at the same price. It reads text and images, writes text, and works across a 1M-token context with up to 128K tokens of output; its knowledge stops at January 2026. Next to Opus 4.6 it sees images in far more detail, follows instructions more literally, and uses a new tokenizer that turns the same text into more tokens.

Anthropic now lists it as a legacy model: still served on its API, the major clouds and in Claude Code, but no longer updated. Anthropic suggests moving to Opus 5.5, which costs less per token.

Is Claude Opus 4.7 still available?

Yes, as a legacy model: Anthropic still serves it but no longer updates it, and it suggests moving to Opus 5.5.

API and clouds

Still on the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS as claude-opus-4-7.

Claude apps

Paid plans can still use it, with a 500K-token context in chat and 1M in Claude Code.

How long

Anthropic will not retire it sooner than a year after its release, and gives at least 60 days' notice before retiring a model.

Successors

Opus 4.8 replaced it at the same price; Opus 5.5 is the current Opus at $4 / $20 per million tokens.

Who is Claude Opus 4.7 best for?

Claude Opus 4.7 is best for teams that already tuned prompts on it, agents that read dense screenshots and diagrams, and hard coding work that needs a pinned model while newer Opus versions are tested.

• Apps tuned for this model

  Teams whose prompts were tuned on Opus 4.7 can keep its model ID while they test a move to Opus 4.8 or Opus 5.5.

• Dense screenshots and diagrams

  It accepts images up to 2,576 pixels on the long edge, more than three times as many pixels as earlier Claude models.

• Hard coding tasks

  Anthropic built it to take on difficult software engineering with less supervision, and it checks its own work before reporting back.

• Multi-session agent work

  It is better than Opus 4.6 at keeping notes in file-based memory, so later tasks need less context up front.

What's included with Claude Opus 4.7?

High-resolution image input, five effort levels, optional thinking, half-price batch jobs, Priority Tier and every major cloud.

High-resolution vision

Images up to 2,576 pixels on the long edge, about 3.75 megapixels, are read at full detail with no setting to change.

Effort levels

Low, medium, high, xhigh and max trade depth against speed and cost; high is the default, and xhigh arrived with this model.

Optional thinking

Thinking stays off unless you turn on adaptive thinking, and requests that force a tool call still work.

Batch API

Half-price input and output for jobs that can wait, with up to 300K output tokens under a beta header.

Priority Tier

Committed capacity with a 99.5% uptime target, which Opus 5 and Opus 5.5 do not offer.

Runs on every major cloud

Available on the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS.

What changed from Opus 4.6?

Better coding and vision, stricter instruction following, and a new tokenizer that turns the same text into more tokens at the same rates.

Coding

SWE-bench Pro rose to 64.3% from 53.4%, and SWE-bench Verified to 87.6% from 80.8%, in Anthropic's tests.

Vision

Images up to 2,576 pixels on the long edge, more than three times the pixels earlier models took.

Tokens

The new tokenizer maps the same text to about 30% more tokens, and the model thinks more at higher effort.

Web research

Its BrowseComp score fell to 79.3% from 83.7%, one of the few tests where it went backwards.

API

A new xhigh effort level, and setting temperature, top_p or top_k away from their defaults now returns an error.

How much does Claude Opus 4.7 cost?

Claude Opus 4.7 costs $5 per million input tokens and $25 per million output tokens on the API, half that on the Batch API, and it is included in Claude Pro from $20 a month; it is not on the Free plan.

Anthropic lists Opus 4.7 as a legacy model; the current Opus 5.5 costs $4 / $20 per million input / output tokens. Opus models are not on Claude's Free plan. Paying yearly brings Pro to $17 a month ($200 up front); Max is billed monthly only. Team seats cost $25 a month (standard) or $125 (premium), or $20 and $100 billed yearly. US-only inference costs 1.1x the standard API rates. Opus 4.7's tokenizer turns the same text into about 30% more tokens than Opus 4.6. It is also sold on Amazon Bedrock, Google Cloud and Microsoft Foundry at those clouds' own prices. Full pricing on claude.com. We checked these prices on 28 Sep 2026.

How do readers rate Claude Opus 4.7?

No one has rated Claude Opus 4.7 yet. Be the first.

Claude Opus 4.7 pros and cons

It improved on Opus 4.6 in coding and vision, but it uses more tokens for the same text, its cyber safeguards block some ordinary work, and newer Opus models now cost the same or less.

What are the best Claude Opus 4.7 alternatives?

Four tools on Recaply do Claude Opus 4.7's job. Ornith-1.0 is open source, so you can run it on your own server.

Claude Opus 4.8Paid

Choose it when you want the same $5 / $25 price with higher scores (69.2% against 64.3% on SWE-bench Pro in Anthropic's table), fast mode and mid-task system messages.

Claude Opus 5.5Paid

Choose it when you can update your code for always-on thinking; it costs $4 / $20 per million tokens and is Anthropic's current default Opus.

Claude Sonnet 5Free

Choose it when price matters more than peak quality; it costs $2 / $10 per million tokens with the same 1M context and the same January 2026 knowledge cutoff.

Ornith-1.0Open Source

Choose it when you want MIT-licensed coding models from 9B to 397B that write their own agent scaffolds.

Questions about Claude Opus 4.7