Anthropic launches Claude Opus 4.7 with stronger coding and new cyber safeguards

Anthropic has released Claude Opus 4.7, a new flagship model that improves coding, vision, and long-running task performance while adding stronger protections for high-risk cybersecurity requests. The company says Opus 4.7 is now generally available and keeps the same pricing as Opus 4.6.

The cybersecurity part makes this release stand out. Anthropic says it is using Opus 4.7 to test stronger safeguards on a broadly available model before expanding access to more capable cyber systems, including Claude Mythos Preview, which it has not released for general use.

Anthropic says Opus 4.7 performs better than Opus 4.6 on advanced software engineering work, with stronger instruction-following, better follow-through on validation steps, and fewer tool errors in longer workflows. Early customer quotes in the announcement also describe gains in coding autonomy, consistency, and error recovery.

What changed in Claude Opus 4.7

Anthropic says the model brings better performance across coding, agents, vision, and complex professional work. In the announcement, the company highlights a 10% to 15% lift on some software engineering tasks, plus better reliability on multi-step jobs.

The company also says vision has improved sharply. One benchmark cited in the launch post shows 98.5% on visual-acuity testing from XBOW, compared with 54.5% for Opus 4.6, which Anthropic uses to support its claim that the new model handles screenshots, interfaces, and diagrams more accurately.

For developers, Anthropic says Opus 4.7 adds a new xhigh effort mode, public beta task budgets for long jobs, and updates inside Claude Code. The company also says the model uses a new tokenizer, which may raise token counts depending on the content and may require teams to retune prompts and budgets when upgrading.

Why the cyber safeguards matter

Anthropic has framed this launch as part of a broader security plan. In Project Glasswing, the company said advanced frontier models have reached a point where they can exceed all but the most skilled humans at finding and exploiting software vulnerabilities. That is one reason Anthropic says it wants stronger controls in place before releasing more capable cyber systems more widely.

The company’s announcement says Opus 4.7 includes stronger safeguards for dangerous cyber misuse while still allowing legitimate security work through controlled pathways. Anthropic also points to its Responsible Scaling Policy and transparency materials, where it says it evaluates frontier systems for catastrophic-risk domains before deployment.

That makes this release more than a normal model refresh. Anthropic is trying to show that cyber capability can ship with governance, not just performance gains, and that live deployment on a public model can help shape how future systems are released more safely. This is an inference from Anthropic’s launch post, Project Glasswing, and its policy framework.

Access, pricing, and rollout

Anthropic says Claude Opus 4.7 is available across Claude products, the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. That gives both direct users and enterprise customers several ways to deploy it.

The company says pricing remains unchanged from Opus 4.6 at $5 per million input tokens and $25 per million output tokens. Keeping pricing flat while adding capability will likely matter to teams already using Opus for coding and agent workflows.

Anthropic’s newsroom lists the Opus 4.7 announcement on April 16, 2026, while the sample article you shared is dated April 17, 2026. The official announcement date is April 16.

Claude Opus 4.7 at a glance

Item	Details
Model	Claude Opus 4.7
Official announcement date	April 16, 2026
Main upgrades	Better coding, stronger vision, improved multi-step reliability
New developer controls	xhigh effort mode, task budgets, Claude Code updates
Pricing	$5 per million input tokens, $25 per million output tokens
Availability	Claude, API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry
Security angle	Stronger safeguards for high-risk cyber misuse
Broader initiative	Project Glasswing

Sources: Anthropic announcement, newsroom, and Project Glasswing pages.

What this means for security teams and developers

• Expect better coding performance and better handling of long, tool-heavy workflows. Anthropic emphasizes reliability and validation improvements in the launch post.
• Watch token budgets after upgrading. Anthropic says the new tokenizer can increase token counts, which could affect cost controls and rate planning.
• Security teams should pay close attention to the safeguard model Anthropic is testing here, because it may shape how future high-end cyber-capable models are released.
• Enterprises that want managed access can use the model through major cloud AI platforms instead of only the direct Anthropic stack.

FAQ