Anthropic's secret model got leaked

Welcome back, Superhuman. Anthropic is apparently finalizing its most powerful model yet, according to a leaked blog post. The rumor alone was enough to send cybersecurity stocks into freefall on Friday. Here’s what we know so far about Mythos — the market-moving new model.

Today: Learn about Claude Mythos, how to add AI voices to any app in 60 secs, and get the latest prompts and trending social posts.

TODAY IN AI

1. Anthropic is working on its next generation of models: An alleged leaked blog post revealed that Anthropic is finalizing its Claude Mythos family of models that also hinted at a new model tier: Capybara. The company expects this to be a “step change” in AI capabilities — claiming it will be the strongest model yet for cybersecurity, coding, and reasoning. The company is testing the model now with no official release date.

2. Gemini makes it easier to switch over from other chatbots: Gemini now lets you upload preferences and chat history from other AI tools, so you can switch to Gemini without losing your existing conversations. To get started just visit Settings in Gemini, click the new “Import” option, and follow the chatbot’s instructions from there. Learn how it works here.

3. OpenAI launches Codex plugins that automate workflows: Codex plugins let users bundle skills, workplace apps, and Model Context Protocol (MCP) servers into reusable workflows. To install one, just open “plugins” in the Codex app and click “install” on your preferred plugin. These updates mirror functionality offered by Claude Code, as OpenAI pivots to compete more directly with Anthropic. Learn more here.

PRESENTED BY WORKOS

Why the same AI prompt gives different answers (and how teams are fixing it)

Same prompt. Same input. Completely different AI output. That’s the messy reality teams face when testing AI—and most are still shipping without solving it.

Nick Nisi from WorkOS shares how they tackled this problem while building AI tools that write code and answer complex questions.

Inside, you’ll learn how to:

• Test AI reliably (even when outputs change every time)

• Catch confident-but-wrong answers before users do

• Build simple systems to measure AI performance

If you’re working with AI in any capacity, this is a practical look at what it actually takes to make it dependable.

Learn how evals work →

FROM THE FRONTIER

The world's best chess players are gaining an edge by selectively ignoring AI — a skill that’s growing in importance

AI in chess. Chess grandmasters have gained surgical precision over their opening sequences thanks to AI tools, often memorizing their first 20+ moves. But when both players prepare with the same tools, they effectively neutralize each other. As a result, draws have surged at the elite level — until recently.

Making the wrong move. Instead of following AI-recommended “perfect” moves, a Bloomberg report shows that champion players deliberately play suboptimal moves. Why? Because it drags opponents into unfamiliar territory where they’re forced to rely on their own critical thinking skills — instead of moves memorized from a computer.

Staying human. As the world grows increasingly reliant on AI tools, knowing when to push back will likely become a competitive advantage. Coaxing the best results from AI is important. But so is challenging its answers, thinking creatively, and not always defaulting to ChatGPT’s suggestion.

Pushing back. Some of life’s biggest wins can come from decisions an algorithm might consider “non-optimal.” Quitting a job. Pivoting careers. Launching a new venture. Pursuing a different business strategy. When making decisions, lean on AI for guidance and ideas — but remember to also trust your creativity, experience, and instincts. At least that’s what the grandmasters are doing.

THE AI ACADEMY

How to add AI voices to any app in 60 secs (No API key required)

• Step 1: Install in one line

  Paste a single curl command into your terminal. No package manager, no dependencies. It takes about 5 seconds.

• Step 2: Authenticate with one click

  Run ‘rime login’ and your browser handles the rest. No API key to find, copy, or store.

• Step 3: Hear your first AI voice

  Run ‘rime hello’ for speech that plays through your speakers immediately. That's the whole setup.

• Step 4: Synthesize anything

  Type any phrase and pick a voice. Run ‘rime tts’ with your text and you'll hear it in under a second. 100+ voices, two models. Arcana for flagship quality, Mist for ultra-low latency.

• Step 5: Take it straight to production

  Run ‘rime curl’ and the CLI outputs a ready-to-use API call. Whatever you just tested in the terminal becomes copy-paste production code. No translation required.

Learn more about the Rime CLI and create your free account ($50 credits included)

PRESENTED BY SPINACH AI

The AI notetaker your agents will love

Your AI tools are only as good as the context you give them. Spinach AI records, transcribes & summarizes your meetings in 100+ languages, then feeds that context to Claude, ChatGPT or Cursor so they can execute. Go from meeting to action, automagically.

Spinach is backed by Y Combinator, Zoom & Atlassian and trusted by over 10,000 organizations.

Try the most accurate & secure notetaker →

IN THE KNOW

What’s trending on socials and headlines today

⌚️ Timesaver: Learn 10 Claude workflows that can save you 10+ hours each week in this viral article — no Terminal or special tools required (1.7M views).

🎬️ Using Seedance 2.0: This post breaks down how you can access China’s viral video generator — before it’s widely released (prompts included).

 📊 Plagiarism Checker: Either Mary Shelley had ChatGPT in 1816 or AI plagiarism checkers still aren’t accurate (8.5M views).

🌀 2AM Flowstate: The co-founder of DeepMind built the perfect background music for optimal focus using Gemini’s new AI music generator.

🥊 LLM Battle: Watch Claude Opus and GPT-5.4 battle it out in a retro-style fighting game.

PRODUCTIVITY

5 New & Trending AI Tools

• 💬 Lindy*: A secure, personal AI agent that runs your work day from iMessage.

• 🍋 Lemon: The AI voice assistant that helps you write 5x faster and respond to emails 12x faster. Just say it and Lemon will write it for you.

• ✨ Decksy: Create researched, professional slides from topics, text, or uploaded documents.

• 📊 Diagrimo: Turn text or ideas into clear diagrams and infographics.

• 🔡 Clico: Select text for answers, summarize articles, and draft replies without leaving the tab.

^{* indicates a promoted tool, if any}

PROMPT STATION

Clarify First Rewrite

Prompt: Rewrite the following text so it’s extremely clear, simple, and easy to understand for someone with no prior context. Remove any ambiguity, break down complex sentences, and make the flow logical from one idea to the next. Keep the original meaning, intent, and key points exactly the same, but improve readability, structure, and coherence so the message feels effortless to follow. Text to rewrite: [paste text].

3D Pop-Out Character

NanoBanana Prompt: {
  "prompt": "A creative 3D illusion artwork of a cute young boy stepping out of a spiral notebook page placed on a wooden desk. The notebook has lined paper with handwritten text at the top. The boy appears in a semi-realistic 3D cartoon style, with soft messy brown hair, natural fresh face, and a cheerful smile. He is wearing a pastel oversized t-shirt, high-waisted light grey jeans, and clean white sneakers. His lower body is sketched in pencil on the notebook page, while his upper body looks fully realistic and popping out of the paper, creating a striking 3D illusion effect. Warm soft lighting, detailed wooden desk texture, shallow depth of field, ultra-detailed, high resolution.",
  "style": "semi-realistic 3D cartoon",
  "lighting": "warm soft lighting",
  "resolution": "high resolution",
  "effect": "3D pop-out illusion with pencil sketch lower body"
}

Grow customers & revenue 

Join companies like Amazon, HubSpot, and Salesforce. Showcase your product to our 1M+ readers and 2M+ followers on socials. Get in touch.

Until next time — Zain, Theodore, & the Superhuman AI team.