← Blog News

Anthropic Launches Claude 4 Sonnet: Desktop Computer Use Out-of-the-Box

By Best AI Tool Editorial Team June 17, 2026 4 min read
Matrix of glowing lines representing computer use code
Share:

⚡ Quick Summary

  • • Anthropic releases Claude 4 Sonnet, setting a new benchmark for developer-focused LLMs.
  • • Introduces desktop-native "Computer Use" capabilities, automating mouse, keyboard, and OS navigation.
  • • Boosts reasoning benchmarks, scoring 91.8% on SWE-bench Verified.
  • • Expands context window to 300,000 tokens with 2x faster output generation.

Anthropic has officially launched **Claude 4 Sonnet**, the first frontier model in its fourth-generation family. Featuring massive upgrades in coding efficiency, logical reasoning, and complex document comprehension, Claude 4 Sonnet's headline capability is **native desktop computer use**—available immediately through the Anthropic API and the updated Claude desktop application.

Native Computer Use: Taking Control of the Desktop

First teased in preview versions of Claude 3.5, native **Computer Use** is now a core feature of Claude 4. The model can view a virtual desktop screen, calculate precise pixel coordinates for clicking buttons, scroll pages, type keystrokes, and navigate complex folder structures in real-time.

In standard tests, Claude 4 Sonnet was tasked with launching a terminal, downloading a dataset, opening Excel, filtering data, generating a chart, copying the image, opening Chrome, and publishing it to a Slack workspace. The model executed the multi-stage plan autonomously with an **89.5% success rate**, outperforming specialized software automation scripts.

Benchmarks: Coding and System Administration

Claude 4 Sonnet marks a massive shift in raw intelligence. On **SWE-bench Verified** (which measures an AI's ability to fix real-world GitHub issues in complex codebases), Claude 4 Sonnet achieved a record-breaking **91.8% score**, solidifying its position as the premier model for software development agents like Cursor and Windsurf.

Additionally, the context window has been expanded to **300,000 tokens** (approx. 225,000 words), allowing developers to upload entire software modules or massive project folders directly into a single prompt session.

Pricing and API Availability

Claude 4 Sonnet is available starting today in the Console API, Amazon Bedrock, and Google Cloud Vertex AI. Anthropic has maintained the pricing structure of Claude 3.5 Sonnet to encourage developer migration:

  • Input Pricing: $3.00 per Million tokens
  • Output Pricing: $15.00 per Million tokens
  • Prompt Caching: Up to 90% discount on repeated prompt inputs.
🎁

Get Our Free AI Tools Guide

Join 50k+ freelancers getting weekly AI tips and tool reviews.

Explore Prompt Library →