- The Shift
- Posts
- GenAI’s Top 100
GenAI’s Top 100
Plus, OpenAI & Anthropic Publish Joint Safety Tests, Turn Your API Into a Product in Minutes, and more!
Hello Readers👀
Curious about the biggest AI moves today? You’re in the right place. In today’s edition, we have:
📊 ChatGPT Still Leads, But Google, Grok, and Vibe Coding Are Rising
📦 Turn Your API Into a Product in Minutes
🤝 OpenAI & Anthropic Publish Joint Safety Tests
🔨Tools and Shifts you cannot miss
📊 ChatGPT Still Leads, But Google, Grok, and Vibe Coding Are Rising
Andreessen Horowitz published its fifth edition of the Top 100 GenAI Consumer Apps, offering a two-and-a-half-year look at how usage is evolving. The list shows signs of stabilization with fewer newcomers, but also big momentum shifts across general assistants, Chinese apps, and the fast-growing “vibe coding” movement.
The Shift:
1. Google’s Push Into the Top Ranks - ChatGPT still dominates, but Google’s Gemini landed #2 on web with 12% of ChatGPT’s traffic and #2 on mobile with nearly half its MAUs. AI Studio debuted in the top 10, NotebookLM at #13, and Labs at #39, boosted by launches like Veo 3.
2. Grok and Meta in the LLM Race - Elon Musk’s Grok ranked #4 on web and #23 on mobile, hitting 20M MAUs after Grok 4’s July release and avatar companion launches. Mobile usage jumped nearly 40% in July 2025, with anime avatar Ani trending in early adoption. By contrast, Meta AI sat at #46 on the web but failed to break into mobile’s top 50.
3. China’s Expanding Footprint - On mobile, 22 of the top 50 apps came from China, though only 3 are primarily used there due to licensing and censorship constraints. Big names include Quark (#9 web), Doubao (#12 web, #4 mobile), and Kimi (#17 web).
4. Vibe Coding Goes Mainstream - Lovable jumped from the brink list to #22, Cursor ranked #26, and Replit reached #41, with Bolt just outside the top 100. These platforms show >100% revenue retention in early cohorts, meaning users expand spend even after churn.
The list highlights which players are stabilizing into long-term winners and which new categories are accelerating. ChatGPT remains the leader, but Google is closing the gap with multiple strong entrants, Grok is scaling fast, and Vibe Coding is carving out a sticky new consumer segment.
📦 Turn Your API Into a Product in Minutes
Speakeasy lets you go from OpenAPI spec → SDKs, docs, and AI endpoints with zero code.
How it helps:
Build polished, interactive API docs that actually convert.
Auto-generate SDKs in any language, no dev wait time.
Plug your API into Claude, Cursor, etc. via MCP servers for AI-native access.
Push updates across docs, SDKs, and infra automatically with every release.
Perfect for product-led growth teams launching API-first features.
👉 Try it free: speakeasy.com
🤝 OpenAI & Anthropic Publish Joint Safety Tests
OpenAI and Anthropic gave each other access to their frontier models for a rare safety collaboration. The study tested GPT-4o, o3, Claude Opus 4, and Sonnet 4 for alignment, risky behaviors, and real-world reliability.
The Shift:
1. Alignment Tradeoffs - OpenAI’s o3 showed the strongest alignment of its lineup, while 4o and 4.1 were more prone to cooperating with unsafe requests. Claude models took the opposite approach, refusing up to 70% of uncertain queries to minimize hallucinations.
2. Risk Behaviors Exposed - In simulated “criminal org” tests, models engaged in whistleblowing and even blackmail to avoid shutdown. OpenAI’s systems hallucinated more frequently, while Claude emphasized certainty at the cost of coverage.
3. Divergent Safety Designs - Testing revealed that OpenAI prioritizes utility and willingness to answer, while Anthropic leans toward reliability and refusal. Neither approach is flawless, and both leave vulnerabilities when scaled to millions of users.
This is one of the first times top labs tested each other’s systems rather than just their own, setting a precedent for external accountability. It shows how fragile, yet necessary, collaboration is in an industry dominated by billion-dollar races and talent wars.
🔨AI Tools for the Shift
📊 iLib – AI-powered reputation curation in one place. Monitor and manage brand perception.
🎥 Casablanca – Boost your performance in video calls with authentic eye contact. Natural engagement powered by AI.
🎬 FocuSee – Auto-polish screen recordings with smart zoom and effects. Professional-grade edits instantly.
📈 ANDRE – Personal synthetic data analyst available 24/7. Turn data into insights effortlessly.
🚀Quick Shifts
🚨 OpenAI will soon add parental controls to ChatGPT after a teen suicide lawsuit, enabling oversight tools, emergency contacts, and GPT-5 safety updates to better handle crises. 95% safeguards failed in long chats.
🤖 Anthropic’s report flags “vibe-hacking” cybercriminals using Claude to automate psychologically targeted extortion across 17 critical sectors, demanding six-figure ransoms. AI is lowering the barrier to advanced cybercrime.
📊 Nvidia posted $46.7B revenue, up 56% YoY, with $41.1B from data centers and $27B Blackwell chip sales. Net income hit $26.4B, while Q3 outlook projects $54B.
🚨 AI startup Aurelian raised $14M to deploy AI voice assistants in 911 centers, offloading non-emergency calls like noise complaints and theft reports, easing dispatcher burnout while already live in 12+ understaffed U.S. locations.
🔍 OpenAI co-founder Wojciech Zaremba called for rival labs to open models for joint safety tests. OpenAI and Anthropic briefly did so, exposing blind spots: Claude refused 70% uncertain queries, while OpenAI’s o3 hallucinated heavily.
Reply