Claude Cowork Goes GA: 5 AI Models & Free Memory in 8 Months

On April 9, 2026, Anthropic's Claude stopped being just a browser tab. Claude Desktop reached general availability (GA — meaning stable release open to all eligible users) with Claude Cowork on macOS and Windows, becoming a true AI automation agent: running tasks in the background while you work, switching contexts across your apps, and scheduling recurring workflows without you watching every step. For anyone who relies on AI automation to get real work done, this is the upgrade that changes the daily routine.

Claude Desktop: From Browser Tab to AI Automation Agent

Claude Cowork launched to general availability on April 9, 2026, for macOS and Windows through the Claude Desktop app. The move from browser to desktop isn't cosmetic — it means Claude can now open local files, run developer tools, and navigate your screen directly, capabilities that a browser-isolated chat window simply cannot deliver.

Three features define the desktop experience:

• Persistent agent threads: Start a long task on desktop, check its progress from your phone via the Claude mobile app (launched as a research preview March 17, 2026)
• Scheduled tasks: Set Claude to handle recurring work automatically — reports, summaries, file organization — without logging in each time (launched February 25, 2026)
• OpenTelemetry monitoring: Enterprise teams receive detailed activity logs showing exactly what Claude did and when, using OpenTelemetry — the industry-standard protocol for tracking software activity across systems

Role-based access controls also arrived with the desktop rollout for Enterprise plans. Admins can now divide users into groups and assign custom capability sets — so a marketing team sees different Claude features than an engineering team, all within the same company account. The Analytics API (the data feed that surfaces engagement and adoption numbers) was added to Cowork simultaneously, giving enterprise operations teams visibility into how Claude is actually being used.

Five Claude AI Models in Eight Months — The Full Upgrade Ladder

Between September 2025 and April 2026, Anthropic shipped five major Claude model versions — roughly one significant upgrade every six weeks. Here's every rung on the ladder:

• Claude Sonnet 4.5 (Sept 29, 2025): Positioned at launch as "best model in the world for real-world agents, coding, and computer use"
• Claude Haiku 4.5 (Oct 15, 2025): The fastest and most cost-efficient option; matches Sonnet 4's performance on coding and computer use at lower operating cost
• Claude Opus 4.5 (Nov 24, 2025): Described at launch as the "most powerful frontier model to date"
• Claude Sonnet 4.6 (Feb 17, 2026): Introduces a 1 million token context window in beta — roughly 750,000 words, or about 12 average novels loaded into a single conversation
• Claude Opus 4.7 (Apr 16, 2026): Targets software engineering and complex long-running coding tasks; adds higher-resolution image analysis compared to previous Opus versions

One friction point in the rollout: Opus 4 and Opus 4.1 were deprecated and removed from the model selector in January 2026 without a transitional period. Developers who had integrated older Opus versions into production pipelines had to upgrade on short notice. Context window compaction (a technique that automatically summarizes older conversation sections to prevent hard cutoffs, enabling indefinitely long conversations) shipped in November 2025 alongside Opus 4.5, which partially softened this disruption.

Claude Design: A Figma Rival Without a Price Tag (Yet)

One day after Opus 4.7 launched — on April 17, 2026 — Anthropic introduced Claude Design under its Anthropic Labs research umbrella. The product lets you collaborate with Claude to create visual outputs: website designs, interactive prototypes, presentation slides, and one-pagers, all within a single AI-powered workflow.

No pricing has been announced. That's a deliberate signal. When a new AI product enters a mature market with no published cost, it typically means the company is absorbing user acquisition expense before repricing once retention is established. The direct targets are Figma (professional design software starting at $15 per seat per month) and Canva (drag-and-drop design at $12.99 per month for Pro). Claude Design collapses the workflow: instead of switching between an AI chat window and a separate design tool, both happen in the same place.

Interactive apps on iOS and Android (launched March 25, 2026) pair with Claude Design to make mobile-first prototyping viable: live charts, diagram sketching, and shareable assets can now be built directly inside a Claude conversation on your phone, without switching to another app. If you're evaluating Claude as a design workflow alternative, the AI for Automation learning guides walk through practical Claude-to-prototype workflows in plain terms.

Memory Goes Free — And Enterprise Gets a Compliance Stack

The single biggest democratization in this eight-month window had nothing to do with model performance: memory went free for everyone.

In March 2026, Anthropic expanded memory access — the ability for Claude to remember facts, preferences, and context from past conversations, so it doesn't start from scratch every session — to all users including the free tier. Previously, memory required a paid plan. The rollout was gradual (Pro plan users received it over a two-week window before free users), hinting at infrastructure scaling constraints behind the scenes. The end state matters: every Claude user, paying nothing or paying full enterprise rates, can now have a persistent AI that learns their preferences over time.

At the enterprise end, eight months of compliance and control additions arrived in quick succession:

• HIPAA-ready Enterprise plans (January 12, 2026): Organizations can now process PHI — Protected Health Information, the legally regulated category covering patient records and medical data — through Claude. This removes a hard legal blocker that previously excluded healthcare companies from using Claude entirely. HIPAA (the US federal Health Insurance Portability and Accountability Act) carries significant fines for breaches involving compliant vendors
• Self-serve Enterprise (February 12, 2026): Companies can now purchase Enterprise plans directly, without involving a sales team — important for procurement speed in organizations with tight budget cycles
• Claude Code in Team plans (January 16, 2026): All Team plan Standard seats now include Claude Code access at no extra cost
• Health and fitness data analysis (January 12, 2026, mobile only): Pro and Max plan users in the US can now connect health data from iOS and Android; requires Android 14 or later on Android devices

The HIPAA certification is a meaningful barrier removed. Healthcare systems, insurers, and digital health startups that were legally blocked from using Claude with patient data — because any breach could trigger regulatory penalties — now have a compliant, auditable path forward.

Three Features Still in "Research Preview" — Eight Months In

Not everything is production-stable. Three major capabilities remain in research preview — a label meaning available but not formally supported, potentially unreliable at scale — well past their initial launch dates:

• Computer use (expanded March 23, 2026, Pro and Max plans): Claude can open files, run developer tools, and navigate screens without manual setup. Still labeled preview after months of access — reliability at scale remains unconfirmed
• Persistent agent threads (launched March 17, 2026): Phone-based management of Claude desktop tasks. The promise is compelling; the preview status signals the experience is still subject to change
• Claude in Chrome (expanded to 10,000 Max users, September 2025): The browser extension that lets Claude work across multiple browser tabs simultaneously. Eight-plus months after its expansion, rollout to additional users and other browsers remains unannounced

For individual users, a "research preview" label is an acceptable trade-off: you get access to genuinely useful capabilities early, with the understanding that behavior may change. For teams building operations on top of Claude, the label means waiting for GA status before deploying in any workflow where reliability is critical.

You can download Claude Desktop and try Cowork now at claude.ai — it's available by default for Pro, Team, and Max plan users. For a plain-language guide on which plan unlocks which features, visit the AI for Automation setup page.

Related Content — Get Started | Guides | More News