Presented by
Good morning, AI enthusiasts. Anthropic just opened its most capable model to everyone. Buried in the model card is a safeguard that can quietly dial back its help if it thinks you're building a rival AI.
Fable 5 tops nearly every benchmark and costs less than half of what Mythos Preview did. So is this the new default model for builders, or the start of a trust problem?
In today's recap:
Anthropic ships Fable 5 with a silent safeguard catch
Gemini 3.5 Live Translate speaks 70+ languages in real time
Set up Claude Fable 5 for agentic coding
Lovable hits $500M ARR on the vibe coding boom
4 new AI tools, prompts, and more
ANTHROPIC
Anthropic launches Claude Fable 5 and Mythos 5
Anthropic
Recaply: Anthropic just launched Claude Fable 5, a Mythos-class model stronger than anything it has publicly shipped, alongside Claude Mythos 5, the same model with some safeguards lifted for vetted cyber defenders.
Key details:
Safety classifiers screen every request. Flagged cybersecurity, biology, and distillation queries get answered by Opus 4.8 instead, and Anthropic says users are told whenever that fallback fires.
Stripe reported Fable 5 finished a 50-million-line Ruby codebase migration in a single day. The same work would have taken a whole team over two months by hand.
The model card admits a separate invisible safeguard that quietly limits help on frontier AI development. Developer Jon Ready's post calling it out passed 500 points on Hacker News within hours.
Pricing lands at $10 per million input tokens and $50 per million output. The model is included on Pro, Max, Team, and Enterprise seat plans through June 22 before shifting to usage credits.
Why it matters: This is the first time a Mythos-class model is open to everyone. The early numbers back the hype, with Fable 5 debuting at #1 on the Artificial Analysis Intelligence Index. But the launch carries a trust problem. Anthropic won't tell users when the rival-model safeguard kicks in. Developers are already asking how to rely on a coding tool that may quietly stop helping them.
PRESENTED BY PITCH
Start with a prompt. Finish with a thinking partner
Recaply: Pitch just launched Pitch Agent, an AI assistant built into your presentation. It generates a full deck from a prompt, then stays in the loop so you can refine slides, pressure-test your story, and draft follow-up content without leaving the editor.
With Pitch Agent, you get:
On-brand deck generation from a prompt, with visuals matched to your template's style
Chat-based edits: rewrite, split, merge, or swap images conversationally
Story Q&A from inside the deck: "What objections am I missing?" "Does this narrative flow?"
Follow-up content on demand: recap emails, talking points, personalized summaries
Gemini 3.5 Live Translate speaks 70+ languages
Recaply: Google just rolled out Gemini 3.5 Live Translate, an audio model that translates speech in near real time across 70+ languages while keeping each speaker's tone, pacing, and pitch.
Key details:
The model translates continuously as someone talks instead of waiting for them to finish. It stays a few seconds behind the speaker and skips the awkward pauses of turn-based systems.
Speech translation in Google Meet jumps from five supported languages to 70+. That unlocks 2,000+ language pairings in one meeting instead of only translating to and from English.
Google says Grab is testing the model for driver and traveler pickup calls, a channel with over 10 million voice calls a month. Every generated audio clip carries a SynthID watermark.
Developers get public preview access now via the Gemini Live API and Google AI Studio. Google Meet starts private preview this month, and the Translate app rollout is global on Android and iOS.
Why it matters: Real-time translation has been a sci-fi promise for decades, and this is the closest a consumer product has come. The new listening mode rolling out on Android lets travelers hold the phone to their ear like a regular call and hear the conversation in their own language. For builders, the Live API turns voice translation into a feature any app can ship, with platforms like LiveKit and Agora already wired in.
GUIDES
Set up Claude Fable 5 for agentic coding
Recaply: In this tutorial, you will learn how to run Claude Fable 5 inside Claude Code and put the launch-day hype to the test with your own one-shot builds.
Step-by-step:
Install Claude Code with
curl -fsSL https://claude.ai/install.sh | bashon Mac or Linux, orbrew install --cask claude-code, then runclaudein a project folder and log in with your Claude account.Run
claude updateto get version 2.1.170 or later; older versions don't show Fable 5 in the model picker. Fable 5 is included on Pro, Max, Team, and Enterprise seat plans at no extra cost through June 22.Switch with
/model fableinside a session, or launch withclaude --model fable. Thebestalias also resolves to Fable 5 whenever your organization has access to it.Describe the outcome, not the steps. Anthropic recommends handing Fable 5 the result you want and letting it plan, so test a one-shot prompt like: "Build a solar system simulator in a single HTML file that derives orbital motion from physics and predicts eclipses."
Expect occasional fallbacks. Requests flagged by the safety classifiers get answered by Opus 4.8 instead, most often on cybersecurity and biology topics, and Claude Code tells you when that happens. Rephrase and retry if a harmless request trips one.
Pro tip: Give Fable 5 ambiguous problems like root-cause debugging and architecture decisions. Anthropic says that's where its extra investigation pays off, and you can skip the usual "verify your work" reminders.
TOGETHER WITH GAUNTLET AI
Gauntlet AI: A fellowship for experienced software engineers
AI changed what senior engineering means. Gauntlet is built for engineers who want to lead that change — not watch it. The most effective way to become AI-first. No tuition. No tutorials. Just the work. Apply now.
Must be a US citizen to qualify.
AI
Lovable hits $500M ARR on vibe coding boom
Recaply: Lovable just crossed $500M in annualized revenue, with 50 million projects built and 720 million monthly visits to apps made on the platform. The vibe coding startup is not yet three years old.
Key details:
Usage has sped up to one million new projects a week. Users are building websites, e-commerce storefronts, and internal tools like CRMs, inventory systems, and HR platforms.
Revenue climbed from the $400M reported in February to $500M now. The company's new Build Economy study finds 35% of builders already earn revenue from what they ship.
According to the study, 80% of Lovable builders come from non-technical backgrounds and 55% have 11+ years of work experience. Founders, designers, and sales pros are the fastest-growing groups.
TechCrunch notes the company was founded in late 2023 and has not yet turned three. The harder test ahead is whether vibe-coded apps survive long-term maintenance.
Why it matters: Vibe coding has been dismissed as a demo trend, and $500M in annualized revenue is a hard number to argue with. The more interesting signal is who's building. Non-technical pros are turning what they already know into products, and many are replacing SaaS subscriptions with internal tools they made themselves. The open question is maintenance. Software breaks constantly, and abandonment rates will tell the real story over time.
TOOLS
Trending AI Tools
⚙️ Claude Fable 5 - Anthropic's most capable model yet, live now in the Claude apps and API, included with paid Claude plans through June 22
🎥 Creatify Agent - Creative agent that turns a product link into finished video ads, now launching campaigns on Meta, TikTok, and Google
🎨 Maket Auto-Complete - Draw a few walls and rooms, and Maket finishes the residential floor plan without moving them
🔊 Gemini 3.5 Live Translate - Google's speech-to-speech translation model covering 70+ languages, rolling out in the Google Translate app
NEWS
What Matters in AI Right Now?
A Munich court ruled that Google's AI Overviews are the company's own statements, not search results, holding Google liable after the feature falsely tied two publishers to scams and rejecting its claim that users should fact-check AI answers.
Cohere launched North Mini Code, its first open-source coding model with 30 billion parameters and 3 billion active under an Apache 2.0 license, scoring a competitive 33.4 on the Artificial Analysis Coding Index.
The New York Times profiled young mathematicians who spent years on hard problems only to watch AI solve them first, asking what that means for careers built on exactly that kind of work.
OpenAI called for an international AI watchdog in a memo from Sam Altman and chief scientist Jakub Pachocki, urging a body that could coordinate "slowing frontier development when needed," one week after Anthropic backed the same idea.
Intel shares jumped as much as 13% after The Information reported Google will order more than 3 million TPUs from Intel in 2028, with Nvidia reportedly testing Intel's tech as TSMC strains to keep up with demand.
Meta launched Workforce Academy, a free five-week program that guarantees graduates a job building its data centers, with $115M behind it this year and pilots in Louisiana, Ohio, Indiana, and Texas.
Nvidia CEO Jensen Huang declined Sen. Elizabeth Warren's invitation to Thursday's hearing on AI, China, and export controls, offering to host lawmakers at Nvidia's Santa Clara headquarters instead; NBC News found no record of him ever testifying before Congress.
The Guardian found that 517 of 809 planned US datacenters sit on land that spent the past year in drought, with large facilities using up to 5M gallons of water a day, the daily use of up to 50,000 people.
