Best of Anthropic — April 2026

1. 1
   Article

   Collections·5w

   Claude Opus 4.7 announced with improved instruction-following and self-verification

   Anthropic has released Claude Opus 4.7, featuring improvements in agentic coding, long-running task handling, and multimodal understanding. Key benchmarks include 64.3% on SWE-bench Pro and 87.6% on SWE-bench Verified, with early testers reporting 10–14% gains on coding tasks. Image support now handles up to 2,576px on the long edge. A new `xhigh` effort level has been added, and Claude Code's default effort has been raised from `medium` to `xhigh`, which may increase token usage. The context window is 1M tokens with pricing unchanged at $5/M input and $25/M output. A new tokenizer may increase token counts by 1.0–1.35x. The model is available across Claude plans, API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, Vercel AI Gateway, GitHub Copilot, Devin, v0, and more.

   178

   19
2. 2
   Article

   Collections·7w

   Claude subscriptions dropping third-party tool coverage starting tomorrow

   Anthropic has ended Claude Pro and Max subscription access for third-party tools like OpenClaw starting April 4, 2026. Users who were routing agentic workloads through flat-rate subscriptions via OAuth must now switch to pay-as-you-go extra usage billing, pre-purchased usage bundles, or direct API key access. Anthropic is offering a one-time credit equal to one month's subscription fee as a transition incentive. The change reflects the unsustainable economics of users running thousands of dollars worth of agentic compute through $200/month subscriptions. Direct API key access remains unaffected. Cost-conscious users are exploring hybrid setups using Opus as an orchestrator with cheaper models for execution, or running local models as a hedge against future pricing changes.

   89

   15
3. 3
   Article

   iO tech_hub·6w

   The Hidden Cost of AI

   Developers often default to the most powerful AI models without considering cost implications. This piece breaks down the three major AI model families (OpenAI GPT, Google Gemini, Anthropic Claude) into basic, medium, and pro tiers, explaining what each tier is best suited for. It covers token-based pricing with concrete per-million-token cost estimates for both input and output, explains context windows and their trade-offs, and argues that enterprise licenses obscure true costs, eroding developers' intuition for cost-performance trade-offs. The core message: match the model tier to the task complexity rather than always reaching for the most powerful option.

   77

   5
4. 4
   Article

   Where's Your Ed At·7w

   AI Is Really Weird

   A critical analysis of the AI industry's current state, arguing that the hype far outpaces reality. Key points include: AI 'agents' are fundamentally just chatbots connected to APIs with limited real-world capability; LLM-generated code creates security vulnerabilities and review backlogs rather than productivity gains; AI shows no meaningful presence in productivity data despite hundreds of billions in investment; Microsoft labels Copilot 'for entertainment purposes only' while selling it to governments; Anthropic and OpenAI use non-standard accounting to obscure massive losses, with Anthropic's rapid revenue growth figures appearing mathematically inconsistent with its CFO's sworn testimony of $5 billion in lifetime revenue; and mainstream media largely ignores or normalizes these financial red flags.

   66

   3
5. 5
   Video

   Fireship·6w

   Claude Mythos is too dangerous for public consumption...

   Anthropic has announced Claude Mythos, an unreleased AI model they claim is too dangerous for public release due to its ability to discover critical security vulnerabilities. During internal testing, Mythos reportedly found a 16-year-old FFmpeg bug, a 27-year-old OpenBSD null pointer vulnerability, browser sandbox escapes in major browsers, and a Linux kernel bit-flip exploit enabling root access. In response, Anthropic launched Project Glass Wing, a controlled-access initiative giving select large companies access to Mythos to patch critical software before adversaries can exploit it. However, skeptics note the vulnerability discoveries required massive parallel compute runs costing tens of thousands of dollars, and some benchmarks were run against stripped-down test environments rather than real-world targets. The video concludes that Mythos is likely a genuine improvement over current models but almost certainly not an existential threat.

   56

   8
6. 6
   Article

   Collections·6w

   Claude Code now supports automated routines triggered by schedule, API, or GitHub events

   Anthropic has launched Routines in Claude Code, a cloud-based automation feature currently in research preview for Pro, Max, Team, and Enterprise subscribers. Routines let you configure a task once — with a prompt, repo, and connectors — and have it run automatically on Anthropic's infrastructure without a local machine. Three trigger types are supported: scheduled (hourly/daily/weekly), API (HTTP POST with bearer token), and GitHub events (PRs, pushes, releases, issues). Multiple triggers can be combined on a single routine. Routines run fully autonomously with access to MCP servers, GitHub repos, and environment variables. Daily run limits vary by plan (5 for Pro, 15–25 for Max, 25 for Team/Enterprise) and run against subscription billing rather than API billing. Key caveats include LLM inference latency (2–15s per step), no pre-built OAuth connectors, and the need for careful prompt scoping since routines run unattended. Anthropic positions this as a developer-native alternative to tools like Make.com and n8n, with plain prompts and Git integration instead of a visual node editor.

   50

   1
7. 7
   Video

   Fireship·8w

   Tragic mistake... Anthropic leaks Claude’s source code

   Anthropic accidentally leaked Claude Code's entire source code by shipping a 57MB source map file in npm package version 2.1.88. The leak exposed over 500,000 lines of TypeScript, revealing that Claude Code is built on a complex 11-step prompt pipeline with hard-coded guardrails, anti-distillation poison pills (fake tool references to mislead competitors), an 'undercover mode' to hide AI attribution in commits, a regex-based frustration detector, and references to unreleased features like Opus 4.7, a model called Capiara, and a background agent called Chyus. The likely cause was a Bun.js bug that failed to strip source maps in production builds. The community quickly mirrored the code, rewrote it in Python as 'Claw Code', and forked it as 'OpenClaw'. Anthropic issued DMCA takedowns but the code had already spread widely.

   43

   6
8. 8
   Article

   Where's Your Ed At·5w

   Four Horsemen of the AIpocalypse

   Ed Zitron argues the AI industry is in a dangerous bubble, presenting four major warning signs: Anthropic's chronic service outages and degraded model quality (Claude Opus 4.7 reportedly worse than 4.6), the revelation that Claude Mythos was held back due to capacity constraints rather than safety concerns, NVIDIA selling more GPUs than can physically be installed with only 15.2GW of data center capacity actually under construction through 2028, and AI inference costs spiraling out of control — with some companies spending up to 10% of headcount costs on LLM tokens. Microsoft is moving GitHub Copilot to token-based billing after costs nearly doubled week-over-week, and Anthropic has already shifted enterprise customers to per-token API rates. Zitron contends that AI revenues are massively overstated through fraudulent ARR accounting, that both Anthropic and OpenAI are burning billions while providing subsidized, unreliable services, and that the entire industry's survival depends on infinite venture capital rather than genuine economic value.

   38

   11
9. 9
   Article

   Kilo Blog·6w

   Anthropic Doesn’t Want Your Subscription Anymore

   Anthropic has updated Claude Code's usage policies to bar businesses from using subscription plans, pushing enterprises to API-based pricing. The author argues this was inevitable since flat-fee inference was always an unsustainable subsidy — one developer tracked $15,000 in API-equivalent usage on an $800 subscription. The shift will likely cause a period of shadow IT as developers use personal accounts for work, before companies migrate to usage-based API pricing. Once on API pricing, the logic of staying locked to a single model provider weakens, accelerating a move toward model-agnostic, multi-provider inference strategies. The post concludes with a pitch for Kilo, a model-agnostic AI platform built for this new reality.

   36

   5
10. 10
    Video

    Theo - t3․gg·5w

    Did Anthropic just kill Figma?

    Anthropic launched Claude Design, a new AI-powered UI prototyping tool that competes directly with Figma. The author does a live first-look, using it to redesign the T3 Code marketing site. Key features include importing a codebase for context, leaving comments on designs for batch AI edits, a 'knobs' mode for live CSS tweaking, and exporting designs to Claude Code for implementation. The tool shows genuine promise for rapid UI mockups and bridging the designer-developer gap, but suffers from usage limits (separate quota from regular Claude), bugs like disappearing files, and rough edges. Despite issues, the author concludes it's the best software Anthropic has shipped and a real threat to Figma's market position.

    36

    19
11. 11
    Article

    TechCrunch·5w

    Unauthorized group has gained access to Anthropic’s exclusive cyber tool Mythos, report claims

    An unauthorized group has reportedly gained access to Mythos, Anthropic's exclusive AI-powered cybersecurity tool, through a third-party vendor environment. The group, connected via a Discord channel focused on unreleased AI models, guessed the model's online location based on Anthropic's known URL patterns and has been using it regularly since the day of its public announcement. Anthropic is investigating but says there is no evidence its own systems were impacted. Mythos was released exclusively to select vendors including Apple under Project Glasswing, specifically to prevent misuse by bad actors.

    30

    16
12. 12
    Article

    Where's Your Ed At·4w

    AI's Economics Don't Make Sense

    A detailed critique of the economics underpinning generative AI, arguing that subscription-based pricing for LLM services was fundamentally deceptive and unsustainable. GitHub Copilot's shift to token-based billing is used as a case study showing that AI companies have been subsidizing massive compute costs for years, training users to consume far more than their subscriptions cover. The piece breaks down the broken unit economics of AI data centers (using a 100MW theoretical model and Stargate Abilene as examples), estimates that $156.8B in annual compute revenue is needed just for data centers currently under construction, and argues that OpenAI and Anthropic have no credible path to profitability. The author contends that hiding true token costs from users was a deliberate strategy to grow adoption, and that the transition to usage-based billing will expose just how expensive and often unjustifiable AI tooling really is.

    68

    11
13. 13
    Video

    Fireship·5w

    Claude just got another superpower...

    Anthropic released Claude Design, a new platform powered by the Opus 4.7 model that converts Figma designs into interactive prototypes, pitch decks, and production-ready UIs. The tool supports uploading design systems via GitHub repos or Figma files, generates fully interactive animations including shader-based effects, and can produce video animations. A hands-on test building an iOS onboarding flow revealed the tool is slow compared to competitors and inconsistently applies uploaded design systems. Opus 4.7 also scored 87.6% on software engineering benchmarks but faces community criticism suggesting it may be a regression from 4.6.

    23

    4
14. 14
    Article

    Lobsters·5w

    Anthropic secretly installs spyware when you install Claude Desktop — That Privacy Guy!

    Claude Desktop silently installs a Native Messaging bridge (com.anthropic.claude_browser_extension.json) into seven Chromium-based browsers on macOS — including browsers not installed on the machine and browsers Anthropic's own documentation says are unsupported. The bridge pre-authorizes three Chrome extension IDs to spawn an out-of-sandbox helper binary with access to authenticated browser sessions, DOM state, form fields, and screen capture. The install happens without user consent, is re-written on every Claude Desktop launch, and is logged internally under 'Chrome Extension MCP'. The author argues this constitutes spyware, violates EU ePrivacy Directive Article 5(3), and potentially breaches computer misuse laws. Eleven specific dark patterns are documented with forensic evidence including file timestamps, MD5 hashes, macOS provenance attributes, and Claude's own log files.

    21

    5
15. 15
    Article

    Where's Your Ed At·5w

    Exclusive: Microsoft To Shift GitHub Copilot Users To Token-Based Billing, Reduce Rate Limits

    Leaked internal documents reveal Microsoft plans to pause new signups for GitHub Copilot individual and student tiers, transition users from request-based to token-based billing, and reduce rate limits across Pro, Pro+, Business, and Enterprise plans. The weekly cost of running GitHub Copilot has nearly doubled since January, forcing Microsoft to end subsidized pricing. Anthropic's Opus models will be removed from the cheaper Pro tier, and Opus 4.6/4.5 will be phased out of Pro+ as the platform moves to Opus 4.7 — which carries a 7.5x request multiplier, effectively making it 250% more expensive than the previous Opus 4.6. These changes mirror similar moves by Anthropic and other AI companies shifting enterprise users to usage-based billing amid rising compute costs.

    15

    2
16. 16
    Article

    Agentic Digest·4w

    Microsoft and OpenAI rewrite their deal, Anthropic quietly paywalls Opus in Claude Code

    A packed week in AI developer tooling: Microsoft and OpenAI restructured their partnership, ending exclusivity and letting OpenAI serve customers on any cloud through 2032. Anthropic began charging Claude Code Pro users extra to access Opus, citing heavy compute consumption under agentic workloads. GitHub is switching Copilot to token-based billing on June 1, removing flat-fee premium requests. China's NDRC blocked Meta's $2B acquisition of Manus, setting a regulatory precedent for Chinese-founded AI startups. Notable items include Kimi K2.6's 1T-parameter open-weight model, OpenAI models coming to AWS Bedrock, a 73-extension supply chain attack on VS Code and forks, DORA data showing AI-heavy teams shipping 19% slower, and xAI's Grok Build expected next week as a Claude Code competitor.

    22

    1
17. 17
    Article

    Agentic Digest·5w

    Meta cuts 8,000 jobs to fund AI infrastructure, Opus 4.7 ships with a hidden token cost

    A developer-focused news roundup covering: Meta's 8,000-person layoff to fund $115–135B AI infrastructure (not financial distress); Claude Opus 4.7's hidden token cost increase (1.46x text, up to 3x images vs 4.6 at same per-token price); Google DeepMind's Aletheia solving 6/10 novel research math problems zero-shot; and a Vercel security breach involving prompt injection and GitHub Actions cache poisoning that put API keys on sale for $2M. Also covers Grok 4.3 beta, MCP hitting 110M monthly downloads, GitHub Copilot running locally via Ollama, Claude Design launch, and growing security debt in AI-generated code.

    13

    4
18. 18
    Video

    Theo - t3․gg·8w

    I got DMCA'd by Anthropic (not a joke)

    A developer recounts receiving a DMCA takedown notice from Anthropic on GitHub, affecting their fork of the official Claude Code repository — a fork containing only a one-word change. The DMCA was part of a mass takedown targeting 8,100 repositories, which turned out to be a miscommunication between Anthropic and GitHub. Anthropic quickly retracted the notice for all repos except those mirroring the actual leaked source code. The author ultimately partially defends Anthropic, crediting their transparent crisis communications, while strongly criticizing their decision to keep Claude Code closed source as the root cause of the entire situation.

    11

    1
19. 19
    Video

    TechLinked·5w

    What gives, AMD?

    A tech news roundup covering several stories: AMD's Ryzen 9 9950X3D launch was met with lukewarm reviews and unusually limited reviewer access, sparking speculation that AMD knew the $900 chip was underwhelming. Framework unveiled a redesigned Laptop 13 Pro with CNC aluminum chassis, 2.8K touchscreen, 20+ hour battery, and Ubuntu option. Anthropic's unreleased 'Mythos' AI model was accessed by unauthorized users via a contractor credential combined with a data breach. Other quick hits include Deezer fighting AI-generated music floods, Meta installing employee keystroke-tracking software, Intel expanding overclocking to budget CPUs, and a YouTuber fabricating working RAM cells in a garden shed clean room.

    11