Skip to content
What are you looking for?. Open searchSR
brands and shortcuts

Sources

The places checked every morning. A primary source publishes first-hand and outranks an intermediary, because retelling loses the nuance.

Primary sources

  • web.dev Blog

    Google's site for web platform guidance, and where Baseline status changes are announced.

  • Cursor Changelog

    Version-by-version record of what changed in the Cursor editor. Short entries, precise dates, no interpretation.

  • MDN Blog

    Mozilla's documentation and explanation of what browsers actually do. For what is arriving in stable right now, the Chrome and WebKit blogs have their own entries.

  • Baseline, zajednički standard četiri pregledača

    Baseline tells you whether a feature is safe for production. It has three states: limited availability means it is not everywhere yet, newly available means it has just landed in every browser, and widely available means it has been present in all of them for thirty months. It replaces the older practice of checking caniuse and hoping for the best.

  • Zvanična dokumentacija za Claude Code

    The documentation moved: old addresses of the form docs.claude.com/en/docs/claude-code/ redirect to code.claude.com/docs/en/, so bookmarks need updating. It covers every environment Claude Code runs in, from the terminal and the VS Code and JetBrains editors to the desktop app and the web, along with a table that points each intent at the right tool.

  • OpenAI News

    OpenAI's own announcements, model cards and pricing changes. A primary source outranks any site that retells it, because retelling loses the nuance and adds excitement the original does not carry.

  • Anthropic News

    The official place where Anthropic publishes everything it ships: models, safety policies, connectors, and research. Reliability is as high as it gets because there is no intermediary, and a new model appears here first.

Other sources

  • Claude Playground

    Anthropic's console tool for trying the Messages API, which replaced Workbench in August 2026. It shows the exact request the SDK would send and the raw response. A pointer, not a citation: the page is an application, so a claim about a parameter is settled in the release notes and only demonstrated here.

  • Google Cloud — Vertex AI pricing

    Google's published price per million tokens for the Gemini family. The ai.google.dev page carries the same numbers but refuses our client, so this is the one we cite.

  • npm registry

    The registry API itself, which answers with the package metadata: every published version and which one is current. The npmjs.com website refuses our client, so this is where a version claim is checked.

  • NoLoJS

    The catalogue itself: patterns that HTML and CSS now handle without JavaScript. Primary for what the catalogue contains, and it is the subject of the piece.

  • llm (Simon Willison)

    The release notes of the tool itself, written by its author. Primary for what changed in which version, and for the dependency that broke.

  • Obsidian Security

    The company's own announcements. Primary for what it says about itself: funding, product, customer numbers — and read as a company saying it, not as an independent finding.

  • Adversa AI

    A security lab that publishes its own attacks against models, with the method and the disclosure timeline attached. Primary for its own research.

  • SecurityWeek

    A security trade desk that reports funding, breaches, and vendor claims with names and figures attached. Citable for what it reports first-hand.

  • 9to5Google

    Android and Google coverage that often carries screenshots and rollout details before the official post. Citable, with the company's own page preferred when it exists.

  • Pew Research Center

    A research centre that publishes its own studies with the method and the sample attached. Primary for any figure that comes from one of its reports.

  • Black Forest Labs

    The FLUX image models, announced and licensed by the lab that makes them.

  • Claude Code Releases (GitHub)

    The release feed for the same tool, useful because it carries dates and diffs that the documentation only summarizes.

  • Claude Code Changelog

    Version-by-version record of Claude Code. Every version number and flag we publish is read from here.

  • OpenAI API Pricing

    OpenAI's published price per million tokens, by model. The only figure we quote for OpenAI pricing.

  • Anthropic Pricing

    The price list itself. Every figure in our model table is checked against this page and carries the date it was checked.

  • Anthropic API Release Notes

    Anthropic's own record of what changed in the API, dated entry by dated entry. Where a claim about a model, a limit, or an endpoint is settled.

  • CSS Working Group, nacrti

    The working group's editor's drafts. This is where a resolution becomes spec text, and it is the primary source for what a feature actually says — as opposed to what a summary of the resolution says.

  • ChatGPT dokumentacija

    OpenAI's product documentation for ChatGPT itself, as opposed to the API docs. For what a ChatGPT feature does and which surfaces it runs on, this is the primary source.

  • arXiv

    Preprint server. A paper here has not been peer reviewed, which is worth saying whenever its numbers are quoted.

  • NVIDIA Technical Blog

    NVIDIA's own developer blog. For results NVIDIA reports about its own systems, this is the primary source, and it usually carries the caveats the coverage drops.

  • Google DeepMind, kartice modela

    The card carries the full benchmark table, the knowledge cutoff and the safety evaluation. Where the launch post and the card disagree, the card is the source we cite.

  • EU Artificial Intelligence Act

    An independent portal that publishes the article text of the AI Act alongside its recitals. Convenient for reading a single article; the Commission pages remain the source for guidelines and dates.

  • Evropska komisija, Digitalna strategija

    The Commission's own policy pages for the AI Act. Guidelines, codes of practice and application dates are published here first, and this is the primary source for all three.

  • Google Chrome blog

    Google's announcement blog for Chrome. Useful for intent and timing; the help pages carry the conditions.

  • Google Chrome pomoć

    Google's own help pages for Chrome. For what a Chrome feature does, who may switch it on and where it stops, this is the primary source rather than the launch post.

  • npm registar

    The package registry. For anything published as a package, the registry entry is the primary source for the version and the documented configuration.

  • Claude dokumentacija

    Anthropic's developer documentation for Claude. The pricing page carries the multipliers most people never read: cache writes, batch, data residency, fast mode.

  • Runway

    Video and image generation studio that also resells third-party models. Its price list doubles as a map of what is current.

  • Luma

    Luma builds the Ray video models and sells them as credits. Its price list spells out the cost per second at each resolution, which makes it comparable with per-second API pricing.

  • Gemini API dokumentacija

    Google's developer documentation for the Gemini API. The pricing page is the primary source for what a model actually costs, updated the day terms change.

  • Ars Technica AI

    Reported technology journalism with a strong security desk. Explains mechanisms, not just outcomes.

  • TechCrunch AI

    Funding, product launches, and company moves. Fast, and worth checking against the primary source.

  • CSS-Tricks

    Practical CSS and front-end technique, with worked examples rather than opinion.

  • Next.js Blog

    Release notes, security advisories, and design decisions for the framework this portal itself runs on.

  • GitHub Changelog: Copilot

    GitHub's running log of Copilot changes, including quiet ones that never get an announcement.

  • Hugging Face Blog

    Model releases, inference benchmarks, and library changes from the largest open model hub. Often the first place a new open model is documented.

  • Google: The Keyword, AI

    Google's own announcements about consumer and developer AI products. Primary source for anything Gemini ships.

  • Smashing Magazine

    Still the best single source that holds design and implementation together. Practical in-depth pieces, with no cookie-cutter tutorials and no reselling of the same content.

  • Simon Willison’s Weblog

    The best single filter for the practical side of working with language models. He tries models the day they ship and writes about tools, prompts, and security holes — prompt injection above all. If you follow one blog, make it this one.