Anthropic releases Claude Opus 4.7, boosting coding and cyber safeguards

Anthropic widened access to Claude Opus 4.7 on Wednesday, putting its newest flagship model into general release across Claude products, the API, Amazon Bedrock, Google Cloud’s Vertex AI and Microsoft Foundry. The company kept pricing unchanged from Opus 4.6 at $5 per million input tokens and $25 per million output tokens, a move that lowers the friction for enterprises weighing whether to swap in the newer system.

Anthropic said the model is stronger than Opus 4.6 in advanced software engineering, long-running tasks, instruction following and vision. In practice, the company is pitching a model that needs less hand-holding on complex coding work, stays steadier on multi-step jobs and handles harder visual inputs such as charts, dense documents and screen interfaces with higher-resolution image input. Anthropic also said testers found the model could catch logical faults during planning and speed up execution, a meaningful claim for teams trying to automate more of the software delivery stack.

The release comes with tighter cyber controls. Anthropic said Opus 4.7 automatically detects and blocks requests that indicate prohibited or high-risk cybersecurity activity. Security professionals with legitimate use cases can apply to a new Cyber Verification Program, a sign that Anthropic is trying to separate defensive work from misuse as it moves more capable systems into broad commercial distribution.

That caution stands in contrast to Claude Mythos Preview, which Anthropic described as its most capable frontier model but not generally available because of its larger capability jump and safety concerns. Anthropic said Opus 4.7 is less broadly capable than Mythos Preview, but still better than Opus 4.6 across a range of benchmarks. The company is using Opus 4.7 as the first model to test new cyber safeguards before any broad release of Mythos-class systems.

The rollout is already spreading through major partners. GitHub said Opus 4.7 is rolling out in Copilot and will replace Opus 4.5 and 4.6 in the model picker for Copilot Pro+ over the coming weeks. Google Cloud said the model is now generally available on Vertex AI and pointed to better ambiguity handling, problem solving, instruction following and high-resolution vision. AWS said Opus 4.7 is available in Bedrock and highlighted stronger agentic coding, knowledge work and long-running task performance.

The broader backdrop is Project Glasswing, launched on April 7, when Anthropic brought in Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA and Palo Alto Networks as partners, alongside more than 40 additional organizations. Anthropic said the initiative includes up to $100 million in usage credits and $4 million in donations for open-source security organizations, underscoring how closely the next phase of the AI race is now tied to enterprise deployment and security controls.