Claude Opus 4.7, Anthropic's latest generally available model, marks a clear step forward in software engineering performance compared to its predecessor, Opus 4.6. The model handles complex, sustained coding workflows with greater consistency, takes instructions with more precision, and introduces self-verification behavior — checking its own outputs before returning results to the user. Early testers report delegating their most demanding engineering tasks to Opus 4.7 with confidence, something that previously required hands-on supervision.

Vision and Professional Output. One of the most concrete improvements is in visual processing. Opus 4.7 handles images up to 2,576 pixels on the long edge — roughly 3.75 megapixels — which is more than three times the resolution supported by earlier Claude models. This enhancement unlocks use cases that require fine visual detail, including computer-use agents parsing dense screenshots and technical diagram analysis. Beyond vision, the model also delivers higher-quality professional documents, interfaces, and presentations, reflecting an improvement in aesthetic judgment for work tasks.

Cybersecurity and Safeguards. Tied to the earlier Project Glasswing announcement, which outlined both the risks and benefits of AI in cybersecurity, Opus 4.7 is being deployed as a testing ground for new protective measures. Its cyber capabilities are deliberately less advanced than those of the more powerful Claude Mythos Preview. Automated safeguards within Opus 4.7 are designed to identify and block requests associated with prohibited or high-risk cybersecurity activity. A new Cyber Verification Program gives legitimate security professionals — those working in vulnerability research, penetration testing, and red-teaming — a path to access the model for authorized work.

Platform Availability and Pricing. Opus 4.7 is live today across all Claude products, the API, Amazon Bedrock, Google Cloud's Vertex AI, and Microsoft Foundry. The model is accessible via the API string claude-opus-4-7. Pricing remains unchanged from Opus 4.6 at $5 per million input tokens and $25 per million output tokens. Developers migrating from Opus 4.6 should note that an updated tokenizer means the same inputs may map to 1.0–1.35 times more tokens depending on content type, and the model generates more output tokens at higher effort levels.

New Features Alongside the Launch. The release comes with several companion updates. A new xhigh effort level offers finer-grained control over the reasoning-versus-latency tradeoff on difficult problems, sitting between the existing high and max settings. Claude Code defaults have been raised to xhigh for all plans. A /ultrareview slash command in Claude Code delivers dedicated review sessions that surface bugs and design issues a careful human reviewer would flag. Task budgets are also entering public beta on the API, giving developers a mechanism to guide how the model allocates token spend across longer runs.

📊 What This Means (Our Analysis)

The cybersecurity safeguard rollout on Opus 4.7 is as significant as any benchmark number. Anthropic is treating this model as a live proving ground — deploying real-world detection systems on a less capable model before extending them to the more powerful Mythos Preview class.

The jump in image resolution support is not a cosmetic update. Processing images at over three times the previous fidelity fundamentally changes what agentic tasks are now viable — from reading complex technical diagrams to pixel-accurate computer-use workflows that were simply out of reach before.

📌 Our Take: What Anthropic is building here goes beyond a single model upgrade — it is laying the infrastructure for deploying more powerful AI safely at scale.