Who built the model?. .
Free AFIRE Guide | AI Academy | Advertise | AI Mastery A-Z
Plus: Qwen 4-Max Is HERE! Impressive Coding, Lower Costs & Frontier-Level Ambitions
Gestalt won 78% of its head-to-head design matchups with Claude Fable 5.2. Now people are guessing whether the mystery model came from Anthropic or China’s Z.ai.
What’s on FIRE 🔥
IN PARTNERSHIP WITH CHATFEATURED
Your buyers ask ChatGPT. That answer has no CPC.
Ads charge per click. Getting cited doesn’t. When buyers ask ChatGPT, Claude or Perplexity what to use, the answer is built from sources, not bids. ChatFeatured shows you which sources AI cites in your category, then writes and publishes the content that gets you cited.
AI INSIGHTS
🕵️ Mystery Stealth AI Sparks Wild Theories After Crushing Fable 5.2?
A mysterious AI model just appeared out of nowhere and CRUSHED Claude Fable 5.2 and even human designers. The craziest part? Nobody knows who’s behind it.
The model, called Gestalt, suddenly took the #1 spot on the Frontier Design Benchmark, launched by Contra Labs and BenchFlow. The results are insane:
-
Gestalt: 1,931 Elo, 90% win rate.
-
Human designers: 1,739 Elo.
-
Claude Fable 5.2: 1,713 Elo.
-
Claude Opus 5.5: 1,691 Elo.
Even crazier, Gestalt beat Fable 5.2 in 78% of their direct matchups. Its designs also showed almost no typical AI-generated design patterns, scoring close to human work on the benchmark’s AI Slop Detector.
Now, the internet is trying to figure out who’s behind it. Some speculate it’s an unreleased Anthropic model, while others suspect a Chinese AI lab like Z.ai. None of these theories have been confirmed.
Who secretly built an AI model that can outperform today’s biggest names in design?
PRESENTED BY HUBSPOT
The Future of Creator Partnerships. Straight From the Creators.
This report delivers the insights you need to build creator relationships that last.
-
What 300+ creators say drives successful, long-term brand partnerships
-
How to build trust from first contact to full commitment
-
What creative freedom actually looks like to the creators
AI SOURCES FROM AI FIRE
1. FREE Gift: Qwen 4-Max Is HERE! Impressive Coding, Lower Costs & Frontier-Level Ambitions. 60% lower average task costs compared to Qwen3.8-Max. Can Qwen4-Max actually compete with GPT, Claude, and the world’s leading AI models?
2. Make Your Claude Apps Look THIS Good with These 5 Ready-to-Use UI Libraries. We’ve picked 5 UI libraries worth exploring to help you build better-looking interfaces with Claude. Less time fixing the design. More time building something you’re proud to share!
TODAY IN AI
AI HIGHLIGHTS
🤖 Nous Research just brought Hermes Agent to Microsoft Teams and Azure. You can ask it to draft a doc or sort your inbox right in Teams, then turn a workflow that works well into a skill your teammates can reuse.
🛰️ Google picked Blue Origin’s New Glenn rocket to launch solar-powered AI satellites in early 2029. Each satellite will carry Google’s own AI chips and pack roughly as much computing power as a data center rack. It’s part of Project Suncatcher.
🧠 Anthropic mathematician Levent Alpöge says Claude helped him find what he believes is a proof of the Riemann hypothesis, a 167-year-old problem about how prime numbers are spread out. The proposed proof still needs to be verified.
🔒 Signal is blocking screenshots on iOS 27 to keep Siri AI from reading private chats through its Onscreen Awareness feature. On Android, the block kicks in when Signal detects an AI tool like Google’s Gemini Intelligence trying to read the screen.
⚡ When ChatGPT slows down, OpenAI uses Codex to figure out why and suggest a fix for engineers to review. Its Performance Factory found the cause of 600+ slowdowns and other performance issues in just over three months.
💰 Big AI Fundraising: BrightAI raised $97M at a reported $750M valuation. Smith Point Capital, founded by former Salesforce co-CEO Keith Block, led the round, with GE Vernova and Khosla Ventures also investing. BrightAI uses AI and sensors to spot problems in power grids, pipelines, and water systems before they break down.
HOT PAPERS OF THE WEEK
1/ miHoYo trains AI agents to play complex games like humans
The miHoYo Honkai AI R&D Team introduces Gamer, an AI agent that plays games using screen images, a keyboard, and a mouse, just like human players. Trained on tens of thousands of hours of gameplay data, it can plan, remember past actions, and complete long tasks across complex 3D games. Key shift: AI agents could learn skills from gaming that help them navigate and interact with real-world environments.
2/ China Telecom trains a 285B AI model on Chinese chips
China Telecom’s Xingchen AGI Lab introduces Xing 4, an AI model family trained using Huawei Ascend chips and MindSpore instead of conventional GPU infrastructure. Its largest reported training system uses 8,192 Ascend chips for a 285B-parameter model, while an open-source 29B model supports 256K-token context. Big impact: China is building more of its own infrastructure for training powerful AI models.
3/ Grid Robotics maps the future of AI-powered robots
Grid Robotics introduces a research roadmap for agentic robotics, inspired by advances from models like GPT-6 Astra. It describes 4 levels of robot autonomy, from picking up objects and completing tasks to managing complex workflows and handling ongoing responsibilities. Big shift: Future robots could move beyond following individual commands toward managing daily routines, remembering preferences, and adapting to changing situations.
NEW EMPOWERED AI TOOLS
-
🎬 Templify AI turns photos and videos into Reels, TikToks, and YouTube Shorts with 1,200+ templates, AI editing, auto captions, and beat-synced transitions.
-
🧪 Plurai Simulation lets Claude Code, Codex, and GitHub Copilot test AI agents in simulated environments to catch bugs before production.
-
🎯 Alsona tracks 32+ buying signals to find high-intent prospects and automate personalized LinkedIn and email outreach to book sales calls.
-
🐳 Docker introduces Docker Sandboxes and AI Governance to help developers build, run, and manage AI agents with stronger isolation and security controls.
-
✈️ Tripmojo is an AI travel agent that helps plan group trips, manage bookings, track flight updates, organize itineraries, and split expenses.
AI BREAKTHROUGH
🤯 GPT-6.2 Astra Cracks 20-Year Math Mystery in 94 Minutes!
GPT-6.2 Astra just solved a 20-year-old math challenge in 94 minutes, using a different approach from the original proof.
Vals AI revealed that OpenAI’s GPT-6.2 Astra had produced what researchers believe is the first computer-verified proof of Smoktunowicz’s Gap Theorem, originally proven by mathematician Agata Smoktunowicz in 2006.
-
94 minutes: Astra completed the proof without human intervention.
-
1,794 lines of code written in Lean, a system that mathematically verifies proofs.
-
$21.58 is the reported cost of the entire run.
-
49/56 problems solved: Astra topped the Autoform-Hard benchmark, ahead of GPT-6.2 Sol and Claude Fable 5.2, which both solved 48.
The craziest part here is that Astra initially failed. After submitting an incomplete proof, it reconsidered its approach and eventually found a different mathematical route that passed verification.
Researchers confirmed the proof with an independent checker and manual review. However, the theorem itself was already known, and the originality of Astra’s approach still deserves further mathematical scrutiny.
Middleman fees are working hard. Just not for you.
Every dollar a platform takes is a dollar that isn’t going toward your TV campaign. Fewer impressions. Less reach. Same invoice.
Universal Ads puts 100% of your budget on the big screen. No platform fees, no managed-service markups, no measurement markups. Just premium, brand-safe TV inventory across 20+ publishers, all through an easy-to-use self-serve ads manager.
New advertisers: use code UALAUNCH26 when you create your account to get $2,500 in matched ad credits when you spend $2,500.
We read your emails, comments, and poll replies daily
Hit reply and say Hello – we’d love to hear from you!
Like what you’re reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team








Leave a Reply