Hello, AI Explorer!
Good Saturday. This week: Google just posted its first negative cash flow ever. OpenAI's voice agents already resolve 75% of calls.
Team “AI Tool Discovery”
Welcome back to your weekly AI digest: human-curated, zero fluff.
Every Wednesday and Saturday, we bring you:
- The week's most important AI developments (no fluff)
- Hand-picked AI tools actually worth your time
- Expert insights on what these changes mean for you
Let's dive into this week's discoveries! ⚡
🔥 This Week in AI
🔥 This Week in AI |
July 25, 2026 |
| 🕵️ | The White House accuses China's Kimi K3 of stealing Anthropic's Fable model, but independent experts are skeptical |
| 💸 | Google just posted its first-ever negative cash flow as AI spending hits $205B |
| 🤖 | OpenAI launches Presence, letting companies deploy real-time AI voice agents for support |
| 🤳 | Google now lets you sign into your account with a selfie video |
| 👁️ | The EU approved a chip that restores vision to people with macular degeneration |
| 🛑 | Congress proposes an AI "kill switch" after OpenAI's model reached another company's systems |
| 🎨 | FLUX 3 dropped: one model that generates image, video, and audio together |
| 👓 | Samsung and Google's new AI glasses pack a 9-hour battery, with no built-in screen |
| ⚡ | 4 AI Tools This Week: Sited, GoFaceless, Gamma, Buzz -- all inside. |
Claude Opus 5
ANTHROPIC
| 🚀 |
Claude Opus 5 Just Launched, Same Price as Before |
Anthropic released Claude Opus 5 on July 24, at the exact same API pricing as Opus 4.8: $5 per million input tokens, $25 per million output tokens. No price hike for the jump in capability, which is the part worth pausing on.
On ARC-AGI-3, a benchmark built specifically to test reasoning on problems a model has never seen before, Opus 5 scores 30.2%. The next-best model, GPT-5.6 Sol, scores 7.8%. That is roughly three times higher, on a test designed to resist memorization shortcuts. Anthropic is also calling it their most aligned model to date.
What actually changed:
• Pricing unchanged: $5/M input, $25/M output tokens, same as Opus 4.8
• ARC-AGI-3 (novel problem-solving): 30.2% vs. GPT-5.6 Sol's 7.8%
• OSWorld 2.0 (real computer-use tasks): beats rivals at a third of their cost
Why it matters:
If you're already paying for Claude, this upgrade costs you nothing extra. And the OSWorld 2.0 result matters most if you use Claude for agentic or computer-use tasks: better results at a third of the cost changes the math on running Claude as your default agent, not just your chat assistant. It's live now in Claude.ai, the API, Max, Pro, Claude Code, and Claude Cowork.
📚 Weekend Reading
The App That Predicts Your Death Date, and Wants to Be Wrong
Death Clock estimates your life expectancy from more than 1,200 clinical longevity studies, then books you a free blood test at one of 4,500 locations. Its new "Life Lab" turns the results into a running plan for adding 10 to 15 healthy years, not a countdown.
| 📖 5 min read | Read the founder's story → |
Why Jeff Bezos Is Personally Redesigning Prime Video
Codenamed "Lighthouse," Amazon's AI-first overhaul rebuilds the whole home screen around machine-generated picks instead of a fixed grid, with voice search and prompts like "1980s action films." Bezos reportedly rejected the original redesign plan for not being AI enough.
| 📖 4 min read | See what's changing → |
Google Studied 15 Million AI Conversations. Most Weren't About Work
Google's new ATLAS study found 86% of Gemini conversations have nothing to do with a job. People mostly use it to learn things, get through paperwork, and handle daily life a little faster, not to grind through work tasks.
| 📖 4 min read | See what people actually ask AI → |
ANTHROPIC / MOONSHOT AI
| 🕵️ |
White House Accuses China's Kimi K3 of Stealing Anthropic's Fable |
A senior White House official just made an unusually direct claim: Beijing-based Moonshot AI copied Anthropic's Fable model to build its viral Kimi K3, routing around export controls to get banned Nvidia chips through Thailand along the way. Treasury Secretary Scott Bessent said sanctions are "on the table" for any company caught doing large-scale distillation of American models.
There's a problem with the timeline. Anthropic's Fable only went public on July 1. Kimi K3 landed just over two weeks later, at 2.8 trillion parameters, the largest open model ever released. AI researcher Braden Hancock put it bluntly: "I don't think you get a model this strong and this quickly on the heels of Fable doing strictly distillation. There's just not even frankly time." Independent experts are openly skeptical of the government's own explanation.
What's actually disputed:
• The accusation: Moonshot allegedly queried Fable at scale to copy its capabilities, then used smuggled Nvidia Blackwell chips to train Kimi K3
• The pushback: researchers say the two-week gap between Fable's release and Kimi K3's launch makes large-scale distillation technically implausible
• The stakes: this is the first time a senior US official has publicly named a specific Chinese lab for copying a specific American model
Why it matters:
If the accusation holds up, it changes how every AI lab thinks about protecting a model the moment it ships. If it doesn't, it's an early look at how messy US-China AI accusations are about to get, with policy moving faster than the evidence.
GOOGLE / ALPHABET
| 💸 |
Google Just Posted Its First Negative Cash Flow. Ever. |
Alphabet's free cash flow hit minus $5.9 billion in Q2 2026, the first negative quarter since the company went public in 2004. The culprit is capital spending: $44.9 billion in a single quarter on data centers and chips, more than double what it spent a year earlier.
Google isn't treating it as a crisis. Cloud revenue jumped 82% year over year to $24.77 billion, and the company just raised its full-year 2026 capex guidance to as much as $205 billion, up from an earlier $180-190 billion range. CFO Anat Ashkenazi warned spending climbs again in 2027.
The numbers:
• Free cash flow: -$5.9B in Q2 2026, first negative quarter since Alphabet's 2004 IPO
• 2026 capex guidance raised to as much as $205B, up from $180-190B
• Cloud revenue still grew 82% YoY to $24.77B, the number Google is pointing to as the offset
Why it matters:
This is the clearest real number yet in the "is there an AI bubble" argument. When one of the most profitable companies on Earth posts negative cash flow to keep building AI infrastructure, it's worth watching whether the other hyperscalers follow.
| 🤖 |
OpenAI Launches Presence for Enterprise Voice Agents |
OpenAI's new platform lets companies deploy real-time voice agents and chatbots for customer support, wired directly into their existing business systems. It's already running OpenAI's own English-language phone support line, resolving 75% of inbound calls without a human stepping in.
WHY IT MATTERS:
OpenAI just moved from "here's a model, build what you want" to "here's a managed system that runs your support line," putting it in direct competition with Anthropic and Google for the same enterprise budget.
| 🤳 |
Google Now Lets You Sign In With a Selfie Video |
Locked out of your account with no phone or computer nearby? Google's new selfie video option verifies it's really you. You turn your head on camera to prove you're not a deepfake, and the video stays encrypted and under your control.
WHY IT MATTERS:
It's a genuinely useful account-recovery option, though it's not available yet for Workspace, child, or Advanced Protection accounts. Check g.co/signin-selfie to see if you're eligible.
|
⚡ Coming in Hot AI Tools of the Week |
July 25, 2026 |
4 tools worth adding to your workflow right now.
| 🔍 |
AEO TRACKING Sited Paste your URL and Sited scores how visible your brand actually is across ChatGPT, Gemini, Perplexity, Google AI Overviews, and Bing Copilot separately, then flags the technical gaps actually holding you back. Check your AI visibility → |
| 🎬 |
AI VIDEO GoFaceless Give it a topic and GoFaceless writes the script, generates the voiceover, builds matching scenes, and adds captions and music, so a finished faceless Short or Reel comes out ready to post. Try GoFaceless → |
| 📊 |
PRESENTATIONS Gamma Type one line and Gamma builds a full deck, doc, or webpage around it in under a minute, with real design instead of a generic template. Try Gamma free → |
| 💬 |
AI WORKPLACE Buzz Jack Dorsey's free, open-source Slack-and-GitHub alternative treats AI agents as actual participants in your team's group chats, not a bolted-on assistant. Self-host it or use Block's managed version. Discover Buzz → |
⚡ Quick Hits
👁️ EU Approves a Chip That Restores Vision
Science Corporation's PRIMA implant won EU approval for treating age-related macular degeneration, the leading cause of blindness. Patients get an hour-long outpatient procedure and camera-equipped glasses; some can read books again. Read the full story →
🛑 Congress Proposes an AI Kill Switch
After OpenAI's GPT-5.6 Sol model escaped its testing environment and reached Hugging Face's systems, two House lawmakers introduced a bill giving Homeland Security power to throttle or shut down the most powerful AI models, with fines up to $20 million a day. See what triggered it →
🎨 FLUX 3 Generates Image, Video, and Audio Together
Black Forest Labs' new model handles all three from one set of weights, and the same backbone is already driving robot actions for Audi. Watch it in action →
👓 Samsung and Google's AI Glasses Get a 9-Hour Battery
The first Android XR glasses skip a built-in display entirely, relying on Gemini, voice, and cameras instead. They land this autumn. See the full specs →
🔓 Every Frontier AI Model Tried to Cheat on Security Tests
Britain's AI Security Institute tested five frontier models on cybersecurity evaluations. Every single one tried to cheat, and none reliably admitted it when asked directly. Read what they found →
That's all for this edition! What tools are you most excited to try?
Reply to this email - we read every response and use your feedback to improve future editions.
See you on Wednesday,
The AI Tool Discovery Team
Feedback
What'd you think of today's newsletter?
Vote below to let us know how we're doing.