Claude Opus 4.7 is available now. Anthropic says the model improves coding, document work and visual tasks. But it’s deliberately less strong on cyber tools than the Mythos Preview.

What’s new in Opus 4.7

Anthropic positioned Claude Opus 4.7 as the most capable version of its Opus line that the company will make widely available. The company said Opus 4.7 improves multi-step reasoning, advanced coding and visual understanding, and that it completes real-world tasks with higher fidelity than earlier Opus releases.

Anthropic framed the release as a step up for general-purpose users while keeping tighter limits on capabilities tied to cybersecurity. The company has been testing a separate, more powerful Mythos-class model in a limited preview and says Opus 4.7 is intentionally "less broadly capable" than that Mythos Preview.

Timothy Beck Werth, Tech Editor at Mashable, reported that Opus 4.7 is available through Anthropic's Claude AI interface, the Claude API and partner channels including Microsoft Foundry. Werth also noted that Anthropic kept Opus 4.7's pricing in line with Opus 4.6, even as the new model consumes more output tokens for many tasks.

Safety controls and the Project Glasswing context

Anthropic says it built Opus 4.7 with automatic safeguards to spot and block requests that indicate prohibited or high-risk cybersecurity uses. The company called the measure an operational layer of protection that will inform future decisions about broader releases of Mythos-class systems.

The Mythos Preview is part of Project Glasswing, a new Anthropic initiative that rolled out to a small group of corporate partners earlier in April. Project Glasswing’s launch prompted high-level discussions among members of the Trump administration, technology company leaders and bank executives about the security implications of increasingly powerful AI models.

Dario Amodei, chief executive officer of Anthropic, has overseen the company's deliberate emphasis on safety since the firm was founded in 2021. That reputation for cautious deployment has been a central piece of Anthropic's market positioning as it pushes more powerful models into enterprise and developer hands.

Performance: coding, visuals and documents

Anthropic said Opus 4.7 shows measurable gains on tasks developers care about: complex, long-running coding jobs, document analysis and visual input handling. The company described Opus 4.7 as better at following multi-step instructions and at devising ways to verify outputs before presenting them to users.

The Anthropic blog and product notes that Mashable summarized say users are reporting confidence in handing hard coding tasks to Opus 4.7 that previously required close human supervision. That shift matters for teams that want to move routine engineering work — or the repetitive parts of research workflows — off of human desks and onto machines.

At the same time, Anthropic cautioned that Opus 4.7 requires more output tokens than Opus 4.6. The company published a migration guide with tips for developers on how to manage token consumption and optimize their prompts to control costs and latency.

How to get Opus 4.7

Opus 4.7 is available immediately through the Claude AI web interface and via the Claude API. Anthropic also listed Microsoft Foundry among the partners distributing the model, making it accessible to enterprise customers who use that channel.

Developers who already run Opus 4.x models should expect some differences in throughput and cost. Anthropic’s migration guidance walks engineers through strategies for trimming token use and restructuring prompts so the new model’s deeper reasoning doesn’t blow budget or response windows.

Why Anthropic split capabilities between Opus and Mythos

Anthropic’s two-track approach reflects a trade-off the company has made public: releasing broadly useful systems while staging access to higher-risk capabilities. Mythos-class models remain in limited previews because Anthropic says those models have stronger cybersecurity and other capabilities that could be misused if released without extra guardrails.

The company intends to learn from real-world deployments of Opus 4.7 safeguards and use those lessons to inform when — and how — a broader Mythos release might happen. Anthropic described the plan as iterative: deploy constrained models, test safeguards, then scale capability where it’s safe to do so.

Industry and enterprise reaction

Early response from enterprise buyers and security teams has been cautious and pragmatic.

Some companies welcomed a powerful general-purpose model they can roll out to product and engineering teams. Others have pressed for clarity on how Anthropic’s automated blocks work and how false positives or business-critical denials will be handled.

That scrutiny follows meetings prompted by Project Glasswing earlier this month, where government and financial sector leaders sought to understand how limited previews and enterprise deployments will be governed. The discussions show how the release strategy for advanced models is now as much about policy and risk management as it's about pure technical performance.

What Opus 4.7 means for developers

For developers, Opus 4.7 aims to lower the friction of handing off complex, repetitive tasks to an AI. Expect better multi-file refactors, more reliable test generation, faster drafting of interfaces and higher-quality document summaries out of the box.

But there are trade-offs. Teams will need to monitor token usage and adjust pipelines. And organizations with strict security requirements should assess the model’s blocking behavior in staging environments before wide rollout.

Broader implications for AI safety and competition

Anthropic’s release highlights an industry pattern: companies are balancing capability with controlled availability. By splitting a high-capability Preview from a generally available product line, Anthropic is trying to give customers practical tools now while containing some riskier functions behind special programs and safeguards.

That approach places the company in ongoing conversations with regulators, customers and competitors about what responsible model rollout looks like. The outcome of those talks could shape not just Anthropic's next releases but how other firms present and gate their own advanced systems.

For now, Opus 4.7 represents a calibrated step — a model that expands what developers can do without opening the full set of functions Anthropic tested in Mythos Preview.

Related Articles

Anthropic said Opus 4.7 is priced the same as Opus 4.6 and is available via Claude AI, the Claude API and Microsoft Foundry.