Toolify Weekly
Hey there, AI builders! 👋 Welcome to another wild week in the AI trenches. If you want a perfect snapshot of the industry's split personality right now, this is it.
On one side of the ring, we have Google throwing open the doors with Gemma 4, putting frontier-level, agentic AI on everything from server racks to your grandma's Android phone. On the exact opposite side, Anthropic just built a model so terrifyingly good at hacking that they’ve locked it in a digital vault. They're calling it a "cybersecurity reckoning." Open innovation vs. existential lockdown—pick your fighter!
Here’s a quick preview of what’s on the menu this week:
Google drops Gemma 4: The most capable open models byte-for-byte, now under an Apache 2.0 license.
Anthropic's Mythos: A model so good at finding zero-day exploits, it's deemed too dangerous for the public.
OpenAI's Profitability Timeline: Spoiler—don't expect them to be in the black until 2030.
Sora is Dead: OpenAI suddenly exits the video generation race to cut costs.
10 Trending AI Tools: From AI accountants to blistering fast TTS APIs.
🔥Top Story #1
The Gist: Google just released Gemma 4, its most intelligent open model family to date. Purpose-built for agentic workflows and advanced reasoning, it’s being released under a commercially permissive Apache 2.0 license.
The Details: Gemma 4 comes in four versatile sizes: Effective 2B (E2B) and 4B (E4B) engineered for mobile-first/edge deployments, alongside a 26B Mixture of Experts (MoE) and a 31B Dense model for heavier lifting. The 31B model is currently punching way above its weight class, ranking #3 on the Arena AI text leaderboard against models 20x its size. The edge models feature a massive 128K context window, while the larger models go up to 256K. Plus, they natively process video, images, and audio out of the box.
Why it matters: Google is putting serious pressure on the open-source ecosystem. By sizing these models to run flawlessly on local hardware—from offline Android phones to a single 80GB H100—and adopting the beloved Apache 2.0 license, they are giving developers the ultimate sandbox. It's a massive win for local-first development, privacy, and highly capable autonomous agents.
🔥Top Story #2
The Gist: Anthropic has developed a new model, Claude Mythos Preview, that is so exceptionally good at identifying zero-day vulnerabilities in software that the company is flat-out refusing to release it to the public.
The Details: Instead of a public launch, Anthropic is sharing Mythos with a consortium of 40+ tech giants (dubbed Project Glasswing)—including Apple, Amazon, Google, and Microsoft—to help them patch their systems. The model has already autonomously found "thousands" of high-severity bugs in every major OS and browser, including a 27-year-old flaw in OpenBSD that humans missed for decades. Anthropic executives are calling this an "industry change point" and a literal "reckoning" for cybersecurity.
Why it matters: We've been warning about AI-powered cyberattacks for years, but this proves the tipping point has arrived. If an AI can autonomously discover decades-old vulnerabilities in minutes, the defensive side of the industry needs a massive head start. It also highlights Anthropic’s unique and somewhat paradoxical position: they are building incredibly powerful AI while simultaneously ringing the loudest alarm bells about its dangers.
Quick hits
News Snacks 🍿
OpenAI Exits Video Generation: OpenAI is abruptly shutting down Sora and exiting the video generation market to reallocate resources to world models and robotics, citing unsustainable costs.
OpenAI Won't Profit Until 2030: Internal documents reveal OpenAI projects a staggering $121 billion spend on computing power by 2028, pushing profitability out to at least the end of the decade.
Gemini API Adds Flex & Priority Tiers: Google is giving developers more control over cost and latency with a 50% cheaper 'Flex' tier for background tasks and a 'Priority' tier for mission-critical apps.
Anthropic Secures Gigawatts of Compute: Anthropic has inked a massive deal with Google and Broadcom for next-generation TPU capacity starting in 2027 to keep pace with its booming $30 billion run-rate revenue.
Claude Code's Source Leaks: An accidental npm publish exposed over 512,000 lines of Claude Code's inner workings, revealing its three-tier memory system and hidden developer features.
Intel Joins Elon Musk's Terafab: Intel is teaming up with SpaceX and Tesla to build a massive new U.S. semiconductor factory in Texas to power future AI and robotics initiatives.
AI Tools
New Tools to Try 🛠️
Jupid AI: An AI accountant for freelancers and LLCs that categorizes expenses and generates your Schedule C directly via WhatsApp or iMessage.
Viktor: A managed AI coworker that lives in your Slack or Teams and natively connects to over 3,000 business tools to pull reports and automate tasks.
Gemma Tuner Multimodal: An open-source toolkit to fine-tune Gemma 4 and 3n with audio, images, and text natively on Apple Silicon.
Hermes Agent: An open-source, autonomous AI agent framework by Nous Research that learns your projects and grows more capable the longer it runs.
NovaVoice: Your 'Voice OS' for desktop, offering lightning-fast smart dictation, formatting, and an AI voice assistant across any app.
Artemis II Lunar Countdown: A sleek macOS menu bar app to track NASA’s Artemis II mission in real-time with live telemetry and event countdowns.
Lessie AI: An AI-powered people search engine that scans 100+ platforms to instantly find influencers, investors, and B2B leads.
Smallest.ai Lightning TTS: A blazing-fast text-to-speech API delivering 100ms latency and 15 languages, built specifically for real-time voice agents.
Pixero: An AI advertising engine with autonomous agents that script, produce, and deliver high-converting video and static ads.
PopTask: A smart, natural-language task manager that lives right in your Mac's menu bar and uses AI to break big tasks into actionable steps.
Stay building, stay curious. See you next week! — The Toolify.AI Team