Claude Opus 4.7 Unleashed: How Anthropic’s New Model Automates Complex Tasks

Claude Opus 4.7 Release Overview

Anthropic has launched Claude Opus 4.7, a large‑language model that can run longer and more complex tasks without continuous human supervision. The update emphasizes autonomous operation, especially when paired with the newly introduced Routines automation feature.

Routines: Bridging Cloud Night‑Shift Workflows

Routines let users configure a single prompt, mount a code repository, and connect multiple API connectors. Once set up, Claude can execute tasks automatically based on schedules, API calls, or GitHub events. For example, the model can wake at 2 a.m. to run a code‑review request, complete testing, locate vulnerabilities, and generate a clean report without any user interaction.

The new /ultrareview command enables a single‑step deep code review, allowing Claude to perform thorough static analysis and produce detailed audit reports.

Significant Capability Enhancements

Across several dimensions, Opus 4.7 shows measurable improvements:

Knowledge‑work productivity is noticeably higher.

Visual input now supports up to 3.75 million pixels—more than three times the previous resolution.

Document reasoning improves by 23.5 %, long‑context reasoning is stronger, and biological reasoning jumps 43.1 %.

Image input now accepts up to 3.75 million pixels, providing richer visual context.

Benchmarking on CursorBench shows Opus 4.6 scoring 58 % and Opus 4.7 jumping to 70 %. Notion’s internal tests report a 14 % overall performance lift and the first successful pass of implicit‑requirement tests, indicating the model now understands unstated engineering conventions.

Safety Mechanisms and Cost Controls

A new guardrail system automatically detects and blocks prohibited or high‑risk network‑security use cases. While the model’s security baseline remains stable, it is deliberately less permissive than the unreleased Mythos Preview, keeping abuse rates low.

Pricing stays at $5 per million input tokens and $25 per million output tokens. The introduction of the xhigh compute tier gives engineers finer control over inference depth versus latency. Claude Code’s default compute level has been upgraded to xhigh, and a new task‑budget feature lets developers cap token spend over long‑running jobs.

Advanced users can enable an automatic mode where the system makes decisions with minimal human interruptions, completing extended tasks autonomously.

The underlying tokenizer has been upgraded, reducing token consumption by roughly 1 × to 1.35 × for the same input while enabling deeper reasoning, which naturally increases output token usage.

Reference

https://www.anthropic.com/news/claude-opus-4-7