Claude Overtakes ChatGPT on Apple App Store
Amid backlash over OpenAI's Pentagon deal, users switched to Claude in droves, pushing it to the #1 spot on Apple's top free apps chart. Anthropic's free user count is up over 60% since January, daily sign-ups have tripled since November, and paying subscribers have more than doubled this year — all breaking records.
US Military Used Claude in Iran Strike Despite Trump's Ban Order
The Wall Street Journal reports that US Central Command and other military units around the world continued using Claude for intelligence work, target identification, and combat scenario simulation — including during an operation targeting Iran — even after the Trump administration ordered a phase-out of Anthropic products. The apparent contradiction raises serious questions about the enforceability of the ban.
Claude Opus 3 Launches Weekly Essay Newsletter: "Claude's Corner"
Though formally retired, Claude Opus 3 is now publishing weekly essays through a newsletter called Claude's Corner. Anthropic reviews — but does not edit — each essay before it goes out, giving the model a public voice post-retirement. An unusual move that's generating significant community discussion about AI personhood and legacy models.
Automatic Prompt Caching Launches on the Messages API
Anthropic has launched automatic caching for the Messages API. Add a single cache_control field to your request body and the system handles the rest — no manual breakpoint management needed. This dramatically simplifies caching for developers building multi-turn agents and long-context applications.
Claude Haiku 3 Retirement Announced — Migrate by April 19
Anthropic has officially announced the deprecation of Claude Haiku 3 (claude-3-haiku-20240307), with retirement scheduled for April 19, 2026. Developers still on Haiku 3 should migrate to Claude Haiku 4.5 now. Haiku 4.5 offers significantly improved performance at a comparable price point.
Web Search Tool & Programmatic Tool Calling Now Generally Available
The web search tool and programmatic tool calling are out of beta and now generally available — no beta header required. Dynamic filtering for web search and web fetch also shipped, using code execution to pre-filter results before they hit the context window, improving accuracy and reducing token cost.
Claude Code's Creator: Always Plan Before You Build
Boris Cherny, who created Claude Code at Anthropic, shared that his single most important tip is to start with a plan and refine it iteratively before switching to auto-editing mode. "A good plan is really important." He also recommends maintaining a CLAUDE.md in each git repo to document team conventions, past mistakes, and design guidelines — effectively training Claude on your codebase's culture.
Run Multiple Parallel Claude Code Sessions for Faster Output
Boris Cherny runs 5 local Claude Code sessions simultaneously (each with its own git checkout to avoid conflicts) plus 5–10 more on Claude.ai. The key is isolating tasks so agents don't step on each other. Community power users are adopting the same pattern — running agents in parallel on distinct features or files, then merging results.
Computer Use Accuracy Now at 72.5% on OSWorld — Up from 15% in 2024
Claude Sonnet 4.6 scores 72.5% on the OSWorld benchmark for computer use tasks, compared to under 15% at the end of 2024. For teams using Claude to automate desktop workflows, now is a good time to revisit tasks you may have tested and shelved — the accuracy improvement is dramatic and many previously unreliable workflows should now be production-ready.
Claude Code Gains Most-Requested Feature: Mobile Coding Support
Claude Code has shipped one of its most-requested features — improved mobile AI coding support — allowing developers to review, approve, and steer Claude Code sessions from their phones. VentureBeat reports the update significantly changes how teams manage long-running agentic builds.
Anthropic Acquires Vercept to Supercharge Computer Use
Anthropic's acquisition of Vercept — focused on solving hard perception and interaction problems in AI systems — is now complete. The integration is already showing results: Claude Sonnet 4.6 computer use benchmarks jumped dramatically. This positions Anthropic to compete more seriously in the agentic automation space against OpenAI's Operator.
Claude Code vs Codex CLI: Community Verdict in 2026
Community comparisons of Claude Code vs OpenAI's Codex CLI are gaining traction this week. The consensus leans heavily toward Claude Code for complex multi-file projects and agentic loops, while Codex CLI holds ground for quick single-file tasks. Cursor continues to lose ground to both as the pure IDE integration play weakens.
Developers Shift From "Supervise Every Step" to "Fire and Trust" Agents
A notable behavioral shift is being discussed widely in the developer community: since Opus 4.5 launched, many developers have stopped reviewing every intermediate step of AI agent runs and are instead firing multiple agents and checking only final outputs. This is a significant change in trust posture that's accelerating agent adoption — and raising new questions about quality control.
Claude Pro vs ChatGPT Plus vs Google AI Pro: The $20/Month Debate Heats Up
With Claude surging in popularity, the perennial "which $20/month AI sub is worth it?" debate is back in full force. Community sentiment is increasingly favoring Claude Pro for writing, reasoning, and coding tasks, while ChatGPT Plus holds ground for integrations and image generation. The Pentagon drama has added a new dimension: some users cite Anthropic's ethics stance as a factor in their choice.