Meta and Google inch toward the Frontier

Hola Decoder😎

If someone forwarded this to you and you want to Decode the power of AI and be limitless, then subscribe now and Join Decode alongside 30k+ code-breakers untangling AI.

🤖 Meta and Google push their AI models closer to the frontier

Meta and Google are upgrading their faster, lower-cost models, with both improving coding, reasoning, and agentic performance while keeping their sights on the industry’s most capable systems.

⚡ Muse Spark Climbs the Rankings

Muse Spark 1.3 Max scored 62 on Artificial Analysis’ Intelligence Index, trailing only Claude Fable 5.1 and Opus 5 despite being significantly cheaper to use.

Zuckerberg said Meta’s larger model, codenamed Watermelon, is “next up,” alongside plans to release the model weights for Muse Spark.

💎 Gemini Flash Gets Smarter Without Getting Pricier

Gemini 3.8 Flash scored 59 on the Intelligence Index with stronger coding, reasoning, and agentic performance while keeping 3.7’s $0.75/$3.75 pricing.

Koray Kavukcuoglu acknowledged Gemini remains “a little below the frontier,” while making clear that reaching the leading edge remains DeepMind’s priority.

Meta and Google are narrowing the capability gap while competing on price and efficiency. With Watermelon still ahead, the next battle is whether these cheaper models can reach frontier performance without frontier-level costs.

Together with Levanta

Your Best Holiday Creators Are Choosing Brands Now

Holiday performance is often decided before the promotion goes live. Creators commit to brands offering attractive economics and strong audience fit, so delayed outreach leaves your team competing for whatever attention remains.

Levanta built The 90-Day Holiday Sprint to turn that shrinking window into measurable momentum. Inside, you’ll learn how to: 

• Set commissions around customer lifetime value, protecting margin while attracting stronger partners.
• Recruit with proven products that give creators better conversion potential.
• Use 30, 60, and 90-day checkpoints to identify winning relationships before seasonal budgets rise.

A 90-day runway gives you time to see which creators convert before holiday budgets climb. That evidence makes every commission increase easier to defend.

Don’t wait until November to build the creator program you will wish you already had. Your 90-day countdown starts now.

✨How to turn any messy prompt into a production-grade one using Prompt Master

Most prompts burn tokens guessing what you meant. Prompt Master fixes that before anything gets sent. Free. MIT licensed. 10K+ GitHub stars.

Step 1: Install - Download the ZIP from github.com/nidhinjs/prompt-master. Open claude.ai, go to Sidebar > Customize > Skills, upload the ZIP. Available in every chat instantly.

Step 2: Drop in your messy prompt - "Turn this into a proper prompt: [your rough idea]"

Prompt Master detects your target tool, picks the right architecture, catches token-wasting patterns, and returns a finished prompt.

Step 3: Use explicit mode for control - "/prompt-master Target: Midjourney Task: Cinematic hero shot of my product"

Step 4: Decompile what isn't working - "Decompile this prompt and tell me what's wrong with it."

Returns a rewritten version with every flaw annotated.

Step 5: Translate across tools - "Translate this Midjourney prompt to Sora."

Prompt Master applies each tool's dialect automatically. One messy sentence in. A production-ready prompt out.

🧠 OpenAI’s Astra technique raises AI monitoring concerns

OpenAI’s upcoming Astra model reportedly uses a technique called “recurrent depth” to improve coding and computer-use capabilities, but the approach is raising questions about whether increasingly capable models will remain monitorable.

The Decode:

Astra reportedly thinks in repeated loops - Recurrent depth lets the model repeatedly analyze the same text before answering, potentially increasing intelligence and performance without requiring a physically larger model.

Performance comes with a monitoring problem - The internal loops reportedly produce mathematical representations rather than readable reasoning, making it harder for researchers to inspect how the model reaches conclusions.

OpenAI reportedly reduced the technique - OpenAI dialed back Astra’s recurrent loops so the model continues producing monitorable reasoning, while promising additional reasoning oversight when the model launches.

OpenAI warns about losing visibility - Chief Scientist Jakub Pachocki said he wants to “prevent a race into unmonitorability,” while acknowledging that current approaches for monitoring model reasoning remain fragile.

Astra highlights a growing tradeoff in frontier AI: techniques that make models more capable could also make their internal reasoning harder to monitor. OpenAI appears to be limiting those gains for now to preserve oversight.

Together with Playbook Pro

Your Growth Strategy Is Costing You. Here’s the Math 

Ad costs are climbing. Most first-time buyers disappear within months, and margins are shrinking fast when discounts drive sales.

We spent 200+ hours on research, distilling what top brands do into a field guide you can read and implement in under 30 minutes.

Inside Rewriting the DTC Rulebook, you’ll learn how to:

📈 Lift retention by 5% to unlock 25–95% profit growth
🔥 Launch emotional campaigns proven to double repeat sales in 8 weeks
📊 Cut paid-ad dependence by 30% in 60 days with the Owned Media Checklist
🧠 Slash decision cycles from 3 weeks to 3 hours using the Assumption Buster Toolkit
💰 Use the Brand Value Canvas to raise prices 15–40% without losing volum

Download the report today and get over 2,000 viral video hooks to fuel your next breakthrough campaign.

🏆 Tools you Cannot Miss:

🌼 Blume – Watches Claude Code, Codex, and Cursor sessions locally, identifies repeated corrections and workflow patterns, then suggests reusable rules, hooks, and skills to improve future coding-agent sessions.

🧠 Causal – Provides an infinite AI-powered canvas for organizing notes, images, files, links, research, and project plans while using the full canvas as context for AI assistance.

📝 Fillo – Lets developers add native forms to their products while handling schemas, validation, file uploads, responses, integrations, and delivery infrastructure behind the scenes.

🌐 Tabbit – An AI-native browser that can understand open pages, work across multiple tabs, fill forms, extract information, and execute multi-step browser tasks through an integrated agent.

🔀 ARBR – Provides a self-hosted control layer for routing AI requests across models while tracking costs, enforcing budgets and guardrails, evaluating outputs, and managing provider fallbacks.

🚀 Quick Hits

🛡️ Amazon is adding an AI-powered scam checker to Alexa for Shopping that verifies whether emails, texts or calls claiming to be from Amazon are genuine by comparing them against the company’s messaging records. 

☁️ Reliance Jio has opened JioPC to all Indian internet users, offering cloud-based virtual PCs that can give computers up to eight years old access to modern AI applications without hardware upgrades. Plans start at ₹1,000 ($11) for two months.

⚖️ The Trump administration has backed OpenAI in its copyright lawsuit with The New York Times, arguing that training AI models on copyrighted text can qualify as fair use and that restricting it could hinder scientific and technological progress. 

⚖️ OpenAI and Sam Altman are facing 30 new lawsuits tied to the Tumbler Ridge school shooting, alleging the company failed to act after ChatGPT conversations about gun violence were flagged. OpenAI says the lawsuits contain false claims about its safety teams 

🎓 New York City is banning classroom AI use for roughly 600,000 public school students through eighth grade for the 2026-27 school year. Teachers will also be barred from using AI to grade assignments, while high schoolers will face limited use. 

🥳Prompt of the day

Create a cozy romantic couple portrait

Image Prompt: Create an ultra-realistic intimate lifestyle photograph of the same young couple from the uploaded reference image, preserving both subjects’ exact facial identities, natural proportions, skin tones, hairstyles, and realistic features.

Context

• Subjects: [Upload couple reference image]
• Setting: Cozy indoor bedroom or study
• Wardrobe: Matching black hoodies
• Mood: Warm, affectionate, relaxed
• Aspect ratio: Vertical 9:16

Instructions

Show the man sitting closely behind the woman with his arms gently wrapped around her while kissing the top of her head. The woman should lean back comfortably against him with her eyes closed and a soft, contented smile. Keep the framing close and natural, with warm window light, soft shadows, realistic skin and hair texture, subtle film grain, and a candid lived-in background with books, photos, or simple room details. Match the original pose, camera angle, intimacy, and composition closely while keeping the scene photorealistic and emotionally natural.

🤳AI Nugget of the Day

Thanks for Decoding with us🥳

Your feedback is the key to our code! Help us elevate your Decode experience by hitting reply and sharing your input on our content and style.

Keep deciphering the AI enigma, and we'll be back with more coded mysteries unraveled just for you!