Claude Opus 4.7

Summary

Claude Opus 4.7 is Anthropic's flagship generally-available model as of mid-2026, released April 16, 2026. It builds on Opus 4.6's 1M-context and Agent Teams foundation by delivering substantial improvements in two specific dimensions: advanced software engineering and vision.

Overview

Claude Opus 4.7 is Anthropic's flagship generally-available model as of mid-2026, released April 16, 2026. It builds on Opus 4.6's 1M-context and Agent Teams foundation by delivering substantial improvements in two specific dimensions: advanced software engineering (with the largest gains on the hardest problems) and vision (higher-resolution image understanding with materially better performance on professional tasks like UI work, slide generation, and document creation). Anthropic positioned the release as a notable jump over Opus 4.6 on coding rigor, instruction-following precision, and self-verification — explicitly framing it as the model users can hand off their hardest coding work to with confidence.

Pricing was held flat at $5/$25 per million input/output tokens — the same as Opus 4.6 — making 4.7 a no-cost upgrade for existing Opus customers. Opus 4.7 is also notable for the cybersecurity safeguards introduced alongside it: the model automatically detects and blocks requests indicating prohibited or high-risk cybersecurity uses, in part to differentiate it from the contemporaneously-released Claude Mythos Preview (which has substantially stronger offensive cybersecurity capabilities and is gated behind Project Glasswing partner access).

Specifications

  • Developer: Anthropic
  • Model String: claude-opus-4-7 (check Anthropic docs for the versioned string)
  • Release Date: April 16, 2026
  • Type: Large Language Model (LLM), Multimodal (text + vision)
  • Context Window: 200,000 tokens standard; up to 1,000,000 tokens via beta header (inherited from Opus 4.6)
  • Access: Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, Claude.ai Pro/Team/Enterprise
  • Pricing: $5 per million input tokens / $25 per million output tokens (unchanged from Opus 4.6)

Capabilities

Advanced Software Engineering: The largest single-version improvement in coding capability since Opus 4.5. Particular gains on the hardest software-engineering tasks — Anthropic positions Opus 4.7 as the model engineering teams can hand off their most difficult work to with confidence. Improved instruction-following and self-verification: the model checks its own outputs more rigorously before reporting back.

Vision: Substantially better image understanding at higher resolution. More tasteful and creative on professional design tasks — produces higher-quality UI mockups, slides, and structured documents than Opus 4.6.

1M Context (inherited): Available via the same beta header introduced for Opus 4.6.

Agent Teams (inherited): Native multi-agent collaboration capability remains supported.

Memory for Managed Agents (Public Beta): Released alongside Opus 4.7 — adds cross-session, filesystem-based agent memory with API control, audit logs, and portable memory stores aimed at enterprise teams building long-running agents.

Cybersecurity Safeguards: Automatically detects and blocks requests indicating prohibited or high-risk cybersecurity uses — a deliberate differentiation from the more permissive (and more capable) Claude Mythos Preview gated behind Project Glasswing.

Limitations

At $5/$25 per million tokens, Opus 4.7 is the same price as Opus 4.6 but remains 5x the cost of Haiku 4.5 and 1.67x the cost of Sonnet 4.6 — for production workflows that don't specifically need Opus's reasoning depth, Sonnet 4.6 (or whatever Sonnet release is current) usually delivers better cost/quality. Cybersecurity safeguards may produce false-positive refusals on legitimate red-team or research workflows; teams with vetted cybersecurity needs are routed toward Mythos Preview via Project Glasswing instead. The 1M context and Agent Teams capabilities remain in beta with potentially higher latency and cost considerations.

Recent Developments

  • April 16, 2026 Launch: Generally available across Claude products, the Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry on day one.
  • Memory for Managed Agents Public Beta: Cross-session, filesystem-based agent memory with API control and audit logs — aimed at enterprise long-running-agent deployments.
  • Industry Context: Released the day after the Pentagon's announcement of seven cleared AI vendors (which excluded Anthropic), and during the same week as White House discussion of pre-release frontier-model review — putting Opus 4.7's safeguards announcement in a politically charged context.
  • Cybersecurity Posture: Released alongside the Mythos Preview / Project Glasswing announcement — Opus 4.7 is positioned as the safeguards-on default while Mythos handles vetted offensive-security partner work.

Last Updated

May 7, 2026

→ Back to Models