Hello, AI Enthusiasts!
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
| ⚡ |
Google's Deep Think obliterates benchmarks—84.6% on ARC-AGI-2, crushing all rivals |
| 🚀 |
xAI restructures post-exodus—Musk reveals lunar AI data center ambitions |
| 🧠 |
China's GLM-5 matches frontier models—open-source at $1/M tokens |
| 💨 |
OpenAI's Codex-Spark hits 1,000+ tokens/second on Cerebras chips |
Google Deep Think reasoning benchmarks
| ⚡ |
Google's Deep Think Obliterates Reasoning Benchmarks |
OpenAI and Anthropic have been grabbing all the 2026 headlines—but Google just reminded everyone why it's still the biggest powerhouse in the AI race.
Google released a major update to its Gemini 3 Deep Think reasoning mode, posting dominant scores across math, coding, and science—while also introducing Aletheia, its Olympiad-level math research agent.
🏆 Benchmark Dominance:
| • |
ARC-AGI-2: 84.6% (crushes Opus 4.6's 68.8% and GPT-5.2's 52.9%) |
| • |
Humanity's Last Exam: 48.4% (new record high) |
| • |
Physics & Chemistry Olympiads: Gold-medal level scores |
| • |
Codeforces: 3,455 Elo (nearly 1,000 points above Opus 4.6) |
🔬 Aletheia Math Agent:
Google also unveiled Aletheia, a math agent that autonomously solves open problems, verifies proofs, and hits new highs across domain benchmarks. The frontier for math and science is quickly moving into uncharted territory.
📱 Availability:
• Live for Google AI Ultra subscribers in the Gemini app
• API access open to researchers via early access program
Why it matters:
After Google dominated benchmarks to close 2025, the focus shifted to Anthropic and OpenAI in 2026. This release reminds everyone that Google is arguably the biggest powerhouse in the AI race. Deep Think's scores are wild, and the frontier is accelerating faster than ever.
| 🚀 |
xAI Restructures Post-Exodus: Musk Reveals Lunar AI Ambitions |
After a wave of departures including key founding team members, Elon Musk's xAI hosted its first all-hands meeting since the SpaceX merger—and posted it online for the world to see.
🏢 New Organizational Structure:
| • |
Grok Team: Chat and voice interfaces |
| • |
Coding Team: Developer-focused products |
| • |
Imagine Team: AI video and image generation |
| • |
Macrohard: AI agents emulating entire companies |
Why it matters:
Musk is no stranger to audacious promises, and his timelines often shift. But by broadcasting xAI's tightened focus and lunar plans, he's making sure the world knows he's aiming to build advanced AI in a way no other giant is—scaling beyond Earth's limits instead of draining them.
Chinese AI GLM-5 model benchmarks
| 🧠 |
China's GLM-5: The New Open-Source King |
China's Z.ai launched GLM-5, a 744B-parameter open-weights model that closes the gap with Western frontier models, sitting just behind Claude Opus 4.6 and GPT-5.2 on Artificial Analysis benchmarks.
📊 Performance Highlights:
• Intelligence Index: 50 (beats Gemini 3 Pro and Grok 4)
• Humanity's Last Exam: 50.4% with tools (beats Opus 4.5, Gemini 3 Pro, GPT-5.2)
• Uses DeepSeek architecture with 40B active parameters
💰 Pricing & Availability:
Open-source MIT license | API at $1/M tokens | Runs on Huawei Ascend chips
WHY IT MATTERS:
Another near-frontier model from China that's knocking at the door. With open weights, competitive pricing, and domestic chip support, the gap with the West is narrowing faster than ever.
| 💨 |
OpenAI's Codex-Spark: 1,000+ Tokens/Second |
OpenAI released GPT-5.3-Codex-Spark, a speed-optimized coding model running on Cerebras hardware, cranking out 1,000+ tokens per second—marking OpenAI's first AI product powered by chips beyond Nvidia.
⚡ The Trade-Off:
Spark trades intelligence for speed—trails full 5.3-Codex on benchmarks but finishes tasks in a fraction of the time. Perfect for quick interactive edits while full Codex handles longer autonomous tasks.
WHY IT MATTERS:
Codex's main criticism has been speed. OpenAI just addressed it while making chip diversification real with the first product built on Cerebras. Real-time coding with instant feedback will change development workflows.
✨ Tool Spotlight
Fireflies
The AI-powered meeting notes assistant trusted by professionals worldwide
If you've ever left a meeting and immediately thought "Wait, what did they say about that deadline?"—or spent 30 minutes re-listening to a recording trying to find one action item—Fireflies is about to save you hours every week.
Fireflies is the leading AI for meeting notes that automatically transcribes and summarizes your meetings with powerful automation features. It transforms meeting audio into searchable, shareable notes with seamless integration for Zoom, Google Meet, and Microsoft Teams.
👥 Perfect for:
| • |
Remote teams juggling multiple meetings daily |
| • |
Professionals who need searchable meeting records |
| • |
Teams collaborating across time zones |
🎯 What makes Fireflies special:
Automatic Transcription & Summarization
Join your meeting, and Fireflies automatically records, transcribes, and summarizes everything. No more frantic note-taking while trying to stay engaged.
Searchable Meeting Library
Find any meeting discussion in seconds. Search by keyword, speaker, or topic across your entire meeting history. Never lose track of important decisions again.
Action Items & Decisions Extraction
Fireflies automatically identifies and extracts action items, decisions, and key points—so you know exactly what needs to happen next without re-listening to the entire call.
Seamless Integration
Works with Zoom, Google Meet, Microsoft Teams. Fireflies joins your meetings automatically, transcribes in real-time, and shares notes with your team immediately after.
💰 Pricing:
Paid subscription required | Contact for custom enterprise pricing
📌 What you should know:
| • |
Integrates with popular conferencing tools (Zoom, Google Meet, Teams) |
| • |
Team collaboration features for sharing transcripts |
| • |
Scheduling and automated reminders included |
💡 Real Impact:
Professionals report getting back 5-10 hours per week by not having to manually take notes, re-listen to recordings, or chase down what was discussed. Your meeting notes are ready before the meeting even ends.
⚡ Quick Hits
💰 Anthropic raises $30B at $380B valuation
Revenue run rate hits $14B with $2.5B from Claude Code alone. Official announcement confirms massive growth trajectory. Read more →
😱 ByteDance launches Seedance 2.0 video model
The viral SOTA video model officially launches with benchmark results and technical blog, but access remains restricted to select users.
💼 Microsoft CEO: White-collar work "fully automated" in 12-18 months
Mustafa Suleyman told FT that Microsoft is pursuing "true self-sufficiency" with AI models for most office tasks.
🔄 OpenAI retires GPT-4o, GPT-4.1, and o4-mini
Older models removed from ChatGPT amid user pushback. Focus shifts entirely to GPT-5 series and specialized models.
⚠️ OAI researcher resigns over ChatGPT ads
Zoë Hitzig warned OpenAI's "archive of human thought" creates "unprecedented potential for manipulation" with advertising model.
📚 Weekend Reading
Google Reminds Everyone Who's Boss in AI Race
Deep Think's benchmark obliteration shows why Google remains the biggest AI powerhouse. With autonomous math agents and Olympiad-level scores, the frontier is moving faster than anyone predicted.
Inside xAI's Post-Exodus Restructure and Moon Ambitions
After losing 10 co-founders and key engineers, Elon Musk broadcasts xAI's new four-team structure and reveals plans for lunar AI satellite factories. Audacious or achievable? You decide.
The China AI Gap Is Closing Faster Than Expected
GLM-5, MiniMax M2.5, and DeepSeek continue pushing Chinese models to frontier levels with open weights and rock-bottom pricing. The competitive landscape is shifting beneath our feet.
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.