Anthropic Releases Claude Opus 4.7 With Vision and Agentic Gains

Anthropic released Claude Opus 4.7 (claude-opus-4-7) on April 16, 2026, upgrading its generally available lineup for complex reasoning, agentic coding, vision-heavy workflows, and memory-driven long-horizon tasks. The model is live across Anthropic products, the public API, and major cloud partners including Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry, and is rolling into GitHub Copilot with a staged rollout and promotional pricing on Copilot deployments.

Claude Opus 4.7 ships with several practical, developer-facing changes. It supports 128k max output tokens and first-class high-resolution image handling up to 2576px / 3.75MP, reducing the need for application-level image scaling and simplifying pixel-coordinate mapping. The release introduces a new xhigh effort level to trade latency and cost for capability on agentic and coding tasks, and a beta task budgets primitive that provides a token-count budget for multi-step agentic loops so the model can prioritize and finish work gracefully. On the API the model ID is claude-opus-4-7.

• •High-resolution image support up to 2576px / 3.75MP, improving localization, counting, and screenshot/document understanding.
• •New xhigh effort level to increase reasoning fidelity for agentic runs.
• •Task budgets (beta) to bound token spend across planning, tool calls, and final output.
• •Continued emphasis on file-system memory and multi-session state for long workflows.

The public API and cloud partners are carrying Opus 4.7 availability. Anthropic kept core API pricing consistent with Opus 4.6 for token rates reported by partners, while GitHub Copilot is applying a 7.5× premium request multiplier as promotional pricing through April 30 for Copilot users. Teams migrating from Opus 4.6 should audit prompt harnesses because Opus 4.7 tends to follow instructions more literally and can change agent behavior.

Anthropic explicitly differentiated Opus 4.7 from its frontier Claude Mythos Preview, describing Opus 4.7 as "less broadly capable" than Mythos but robust for production use. The company deployed automated cybersecurity safeguards that block or flag high-risk requests and is using Opus 4.7 to test mitigation strategies before attempting wider Mythos-class releases. As Anthropic put it, "We are releasing Opus 4.7 with safeguards that automatically detect and block requests that indicate prohibited or high-risk cybersecurity uses," positioning Opus 4.7 as a safety iteration point.

Opus 4.7 is a pragmatic release that pushes multimodal and agentic capabilities into broadly available channels rather than the more restricted Mythos track. The larger image resolution, longer token horizons, and explicit tooling for agent budgets reflect production realities: engineers are building longer, autonomous workflows and need models that can reason, call tools, and manage output size predictably. For practitioners, the changes reduce engineering friction around screenshot-heavy tasks and create new knobs for balancing cost, latency, and capability.

Monitor prompt and harness regressions as Opus 4.7 applies instruction-following behavior more strictly, validate token-cost impact from high-res images, and track Anthropic's safety telemetry as it uses Opus 4.7 to iterate controls ahead of any broader Mythos deployment.