Anthropic Claude Opus 4.7: 7 Critical Facts About the New Flagship Model

1. Anthropic Claude Opus 4.7 is Anthropic’s most capable generally available model, but not its absolute most powerful preview model

The first thing to understand about Anthropic Claude Opus 4.7 is the positioning.

Anthropic says Opus 4.7 is its most capable generally available model. That wording matters. In the same launch materials, Anthropic also says Mythos Preview remains more broadly capable and better aligned overall, but Mythos is still limited rather than broadly released.

So Opus 4.7 is the new flagship for broad use, not the ceiling of everything Anthropic has in research preview.

2. Anthropic Claude Opus 4.7 keeps Opus 4.6 pricing while expanding across Anthropic’s full product stack

The model is available today across Claude’s paid products and across the main developer platforms Anthropic supports.

According to Anthropic’s official pages, it is available for Pro, Max, Team, and Enterprise users in Claude, and for developers through the Claude Platform, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.

The pricing point stayed the same as Opus 4.6. It costs $5 per million input tokens and $25 per million output tokens. That same-price upgrade story is one of the strongest commercial angles in the launch.

3. Anthropic Claude Opus 4.7 is mainly a coding and agentic-work release

The release is being sold first as a model for hard software and agent workflows.

Anthropic says the model is a notable improvement on Opus 4.6 in advanced software engineering, especially on difficult long-running tasks. The company and its early-access partners cite a range of gains, including a 13% lift on a 93-task coding benchmark, stronger results on CursorBench, better task completion on Rakuten-SWE-Bench, and fewer tool errors in longer multi-step workflows.

Those are mostly Anthropic and partner-reported evaluations, so they should be read carefully. But the directional signal is clear. Opus 4.7 is being optimised for production coding, long-horizon planning, and agent reliability more than for casual chat.

4. Anthropic Claude Opus 4.7 brings the biggest vision upgrade in the Claude line so far

One of the most important non-coding changes in Opus 4.7 is vision.

Anthropic says the model is the first Claude model with high-resolution image support. The maximum image resolution increases from 1568 pixels on the long edge to 2576 pixels, or roughly 3.75 megapixels. Anthropic also says image coordinates now map 1:1 with actual pixels, which simplifies screenshot and computer-use workflows.

That is a real upgrade for diagrams, dense screenshots, document analysis, and visual tool use. But it also makes image-heavy workloads more expensive, because full-resolution inputs can use up to about three times as many image tokens as prior models.

5. Anthropic Claude Opus 4.7 introduces new control levers for intelligence, effort, and long-run budgeting

This is not just a smarter model. It also changes how developers are expected to control it.

Anthropic added a new xhigh effort level between high and max, and the docs recommend high or xhigh for coding and agentic use cases. It also launches task budgets in beta, which let teams give the model an advisory token budget across a full agentic loop instead of only setting a hard max_tokens ceiling.

It still supports adaptive thinking, but adaptive thinking is off by default unless you explicitly enable it. That is an important detail for teams expecting thinking behaviour to appear automatically.

6. Anthropic Claude Opus 4.7 has real migration-breaking changes that developers cannot treat as minor

This is the part many launch summaries skip.

The model removes extended thinking budgets, so the old thinking: {type: "enabled", budget_tokens: N} pattern now returns a 400 error. It also rejects non-default temperature, top_p, and top_k values, omits thinking text by default unless you opt in, and uses a new tokenizer that may consume roughly 1.0x to 1.35x as many text tokens as Opus 4.6.

Anthropic also says assistant-message prefills remain unsupported on this model, just as they were on Opus 4.6. In practical terms, it is a better model, but it is not a drop-in replacement if your product depends on old reasoning settings, token assumptions, visible thinking traces, or custom sampling parameters.

7. Anthropic Claude Opus 4.7 is also part of Anthropic’s staged cyber-safety rollout

The release matters for safety policy as well as product performance.

Anthropic says the model is the first less-capable model on which it is deploying the new safeguards that automatically detect and block prohibited or high-risk cybersecurity requests. The company says what it learns from real-world deployment of those safeguards will help it move toward broader release of Mythos-class models later.

Anthropic’s safety write-up says Opus 4.7 is modestly better than Opus 4.6 on some measures such as honesty and prompt-injection resistance, modestly weaker on some others, and overall similar in safety profile. That makes the launch more nuanced than a simple “bigger and better” story.