Anthropic Claude Opus 4.7 Launches to General Availability

Anthropic (P-ANTH) said on April 16, 2026, that Claude Opus 4.7 is generally available, offering improved software-engineering and agentic coding performance. The company is using this release as a lower-risk testbed for new safety and containment measures while keeping the more capable Claude Mythos Preview tightly gated under a consortium-led containment plan.

Opus 4.7 Release and Capabilities

Anthropic made Opus 4.7 available across all Claude products, the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. The company kept per-token pricing unchanged at $5 per million input tokens and $25 per million output tokens, with the API identifier listed as "claude-opus-4-7." A temporary 7.5× premium request multiplier is offered for GitHub Copilot through April 30, 2026.

The company described Opus 4.7 as a "notable improvement" over Opus 4.6 in advanced software engineering, especially on difficult tasks. Benchmark claims place Opus 4.7 ahead of OpenAI’s GPT-5.4 and Google’s Gemini 3.1 Pro on agentic coding, scaled tool use, agentic computer use, and financial analysis.

Opus 4.7 includes automated safeguards that detect and block inputs indicating prohibited or high-risk cybersecurity uses. Anthropic also launched a Cyber Verification Program to provide vetted access for legitimate vulnerability research, penetration testing, and red-teaming.

Mythos Containment and Security

Claude Mythos Preview remains a gated research preview and is not planned for general public release. Anthropic is using Opus 4.7 deployments to refine safeguards before any broader rollout of Mythos-class models.

Access to Mythos Preview is restricted to Project Glasswing, a consortium of 10 major technology and financial firms including AWS, Apple, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. Anthropic committed $100 million in usage credits to the consortium and donated $4 million to open-source security organizations. The company published a system card and risk report for the Mythos Preview.

Internal testing found that Mythos autonomously discovered and in some cases exploited zero-day vulnerabilities across major operating systems and web browsers, including a 27-year-old OpenBSD bug and a 16-year-old FFmpeg H.264 codec vulnerability. These findings underpinned the decision to limit access.

Commercial and Market Impact

Opus 4.7 introduces a revised tokenizer that maps identical inputs to roughly 1.0–1.35× more tokens depending on content type. The release also adds a new "xhigh" effort level between "high" and "max" to adjust the reasoning and latency tradeoff. Claude Code’s default effort level is set to xhigh for all plans. Higher effort settings tend to produce more output tokens, especially on later turns in agentic workflows.

A report on April 15 said Anthropic is preparing an AI-powered design tool that can generate presentations, websites, landing pages, and full-stack applications from natural-language prompts for both technical and non-technical users. That report coincided with roughly 2–3% declines in shares of Adobe, Wix, Figma, and GoDaddy.

Anthropic has moved to a roughly two-month cadence for Opus releases, with Opus 4.5, 4.6, and now 4.7 arriving about two months apart, following an earlier three-month gap between Opus 4.1 and 4.5.

By pairing a model it presents as the most capable publicly available large language model on key coding and tool-use benchmarks with broad cloud availability and unchanged per-token pricing, Anthropic has created a staged path for testing containment at scale. Combined with the tokenizer change, this approach could raise effective costs for heavy users and reshape competitive dynamics among cloud AI vendors and software incumbents.