Anthropic moves Computer Use, Skills API, Files API and Browser Use to general availability
On Aug 20 Anthropic pushed four agent capabilities to GA on Claude Platform: Computer Use supports multi-action turns, Browser Use targets DOM rather than pixels, Skills API provides versioned team workflows, and Files API offers 1TB per-org storage with 5x rate limit; a medical claims case dropped from 32 to 13 minutes.
On August 20 Anthropic moved four agent capabilities on the Claude Platform to general availability: Computer Use, Skills API, Files API, and the new Browser Use tool. The goal is to extend Claude from "calling structured APIs" to "operating the full software stack."
Computer Use now uses the computer_toolset_20260801 toolset, allowing Claude to return several sequential actions (click, type, screenshot, etc.) inside one model turn, cutting model call count and latency. In a published medical-claims case the task dropped from 32 to 13 minutes, with model calls reduced 32 to 52 percent, per-task cost down 25 to 32 percent, and completion rising from 77 to 100 percent. The capability is HIPAA-eligible under a BAA, suitable for regulated healthcare workloads.
The new Browser Use tool (browser_toolset_20260801) replaces fragile pixel-coordinate targeting with screenshot plus page accessibility structure, element references, form controls, tabs, and navigation — directly identifying inputs and buttons with far higher reliability than pure pixel clicks. Skills API lets teams upload and version folders of instructions, scripts, and templates; the skill name and short description stay in context (~100 tokens), while the full SKILL.md loads only when triggered and scripts run in a bash sandbox consuming only their output for tokens. Versions are pinned via version_id or the latest keyword. Files API provides 1 TB per organization with auto-expiration and a rate limit of 500 RPM, 5x the previous cap.
All four capabilities also shipped to Microsoft Foundry; Google Cloud Vertex AI support is in progress. Anthropic simultaneously opened Claude Academy, a free learning hub with curricula from Claude 101 to advanced Claude Code engineering. Docs recommend deploying these capabilities under a BAA, isolating browser/desktop agents in containers, restricting to domain allowlists, applying least privilege, and requiring human confirmation at consequential steps.