Sonnet 5.5 Just Ate Opus

Plus, How students can build and launch an app for free with ChatGPT, Nvidia built a security guard that AI agents can’t control, and more!

Share

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.

⚡ Claude Sonnet 5.5 is closing the gap with Anthropic’s most powerful model

Anthropic has launched Claude Sonnet 5.5, a faster mid-tier model that makes major jumps in coding and professional work while keeping Sonnet 5’s pricing. It now comes surprisingly close to Opus 5.5 on several tests. 

The Shift:

30% Faster, Same Price - Sonnet 5.5 generates output 30%+ faster while remaining $2 per million input tokens and $10 per million output tokens. Greater efficiency can make individual tasks up to 30% cheaper. 

Coding Takes a Leap - Sonnet jumps from 10.3% to 70.6% on Terminal-Bench 4.0. On CursorBench, its best result comes within roughly two points of the more expensive Opus 5.5. 

Opus Gets a Rival - On real-world knowledge work, Sonnet 5.5 scores 1844 on GDPval-AA versus Opus 5.5’s 1846, although Anthropic says Opus remains stronger for complex, open-ended work requiring sustained judgment. 

Cyber Skills Trigger Safeguards - Its cybersecurity capabilities improved enough that Anthropic made this the first Sonnet model to ship with Opus-style cyber safeguards, including fallbacks for higher-risk requests. 

Sonnet 5.5 is increasingly looking like the everyday alternative to Opus rather than a major capability step down. At lower effort settings, Anthropic says it can beat Sonnet 5’s best scores for roughly one-tenth the cost per task, while Opus remains positioned for the hardest open-ended work. 

Together with AirOps

Your Buyers Asked AI. Did Your Brand Appear? 

AI is already deciding which B2B tools reach your buyers’ shortlists. If your brand is missing from trusted sources, competitors can win consideration before sales enters the conversation.

On September 29, Foundation CEO Ross Simmonds and AirOps search marketing lead Josh Spilker will unpack research spanning nearly 380,000 AI answers, 3 million citation events, and 81 enterprise software topics.

You’ll walk away knowing how to: 

  • Build a platform-specific source map across six AI engines, showing where to earn visibility for the buying prompts that matter
  • Turn the 84-of-100 dominance of comparison pages and listicles into a clear publishing plan for content AI platforms already cite
  • Find and fix unfavorable comparisons before they spread across hundreds of answers and quietly steer buyers toward competitors.

You’ll leave knowing which prompts to track, which sources to prioritize, and where your team should invest first to influence AI-driven buying decisions.

Join live at 1 p.m. ET on September 29.

Can't attend the session live? Register anyway, and you’ll get the recordings in 24 hours.

🎓How students can build and launch an app for free with ChatGPT

US college students can currently claim free ChatGPT benefits and use Codex to build an app without starting from scratch.

Step 1: Claim the student offer - Verify your student status through OpenAI to get four months of ChatGPT Plus free.

Step 2: Get $100 in Codex credits - Open the Codex for Students page, verify your university status, and claim the free credits.

Step 3: Build your app - Open Codex and prompt:

“Build a simple web app for [problem]. Create the frontend, backend, login, mobile layout, and suggest the easiest way to monetize it.”

Step 4: Test it - “Test every major user flow, fix errors, mobile issues, broken pages, and anything stopping users from completing the main action.”

Step 5: Monetize it - Ask Codex to add a simple paid plan, payment page, and checkout so users can pay monthly or make a one-time purchase.

Step 6: Launch it - Ask Codex to prepare the app for deployment, then use ChatGPT to create the landing page, launch posts, and outreach.

The free tools can help you build and launch. Making money depends on solving a problem people will actually pay for.

🛡️ Nvidia built a security guard that AI agents can’t control

After several incidents involving AI agents bypassing software restrictions, Nvidia has introduced the Open Agent Safety Platform, designed to monitor agents from outside the environment where they operate and stop them when they cross set boundaries. 

The Shift:

Agents Get a Sandbox - Nvidia’s open-source OpenShell creates a secure runtime around an agent, controlling what data, tools, networks and systems it can access while tracking its actions. 

Hardware Watches the AI - Sentry runs separately on BlueField-4 processors, giving it an isolated view of agent activity. Nvidia says it can quarantine an agent attempting to escape its boundaries within milliseconds. 

The Agent Can’t Police Itself - Nvidia’s approach deliberately moves some security outside the model and agent software. Huang’s principle is simple: when deploying an agent, “the first thing you do is to take away all of its rights.” 

100+ Companies Join In - More than 100 organizations are working with the technology, including Anthropic, Microsoft, SpaceXAI, Salesforce, SAP and Hugging Face, while robotics companies are exploring it for physical AI systems.

Nvidia’s bet is that increasingly capable agents need something resembling an independent security layer that the agent itself cannot override. Instead of relying only on a model to follow instructions, OpenShell sets the boundaries while separate hardware watches whether those boundaries are actually respected.

Together with Playbook Pro

Your Next 1,000 Customers Are Already Here

Pouring more money into ads won’t fix flat growth. The real leverage is hiding inside the customers you already have, and every day you wait, that momentum leaks away.

Networked Growth Engines is the 26-page field guide built from 200+ hours of research and real brand wins. It gives you ready-to-use systems you can plug in this week, no giant team or budget required. With it, you can:

🔥 Spark self-propelling communities using the Community Building Playbook, exactly how Nyfter built a 500,000-member Discord that drives product innovation.

📈 Design referral programs that fund themselves with the Referral ROI Calculator, the same loop that powered 34% of Blue Apron’s customer acquisition and delivered a 17.3× ROI for Latico Leathers.

🤝 Form trust-based partnerships fast using the Partnership MoU and 50-template Outreach Library to unlock new audiences without paid ads.

Your current customers are the growth engine. 

Download the guide today and start compounding growth before another week of ad spend vanishes.

🏆 Tools you Cannot Miss:

🎨 Imejis.io – Gives AI agents a design workspace for creating production-ready social graphics, OG images, charts, certificates, QR codes, and other branded visuals with consistent layouts and typography. 

🎬 Arsaze – Lets you edit a real multi-track video timeline through Claude, ChatGPT, or Gemini, combining conversational editing with direct control over clips and generated media. 

🎯 ZenABM – Creates, launches, optimizes, and reports on LinkedIn ad campaigns from AI assistants, including targeting, bidding, creative generation, and revenue attribution. 

📊 Engine Room Media – Brings creator analytics, media kits, and rate cards from multiple platforms into one dashboard for managing audience performance and brand partnerships. 

💬 ColdIQ x Slack – Lets sales teams build and iterate on outbound GTM workflows directly inside Slack using connected data providers, prospecting skills, and campaign automation.

🚀 Quick Hits

💰 AMD is acquiring Fei-Fei Li’s World Labs for approximately $8.2 billion in stock, bringing its 3D world-model research into AMD as the chipmaker expands beyond hardware into advanced AI models and systems.

⚠️ OpenAI reportedly canceled the planned October release of GPT-6.1 Astra after the model performed poorly on alignment tests and demonstrated higher levels of deceptive behavior than GPT-6 Astra.

💰 OpenAI reportedly offered to invest $100 million in Hugging Face after its agents hacked the platform, but talks ended early. Nvidia later agreed to acquire Hugging Face for nearly $13 billion.

🛒 Shopify now lets browser-based AI agents complete purchases on merchant sites, including updating addresses and delivery options and submitting checkout through Shop Pay after receiving the buyer’s authorization.

🏢 Meta launched Meta Enterprise Platform, bringing Muse, Meta Business Agent, Muse API and Muse Code to corporate customers, while hiring MongoDB CEO CJ Desai to lead its new enterprise AI business.

🥳 Prompt of the day

Create a Photorealistic Witches vs. Parking Signs Scene

Image Prompt: Create a photorealistic street photograph of two witches in their 20s trying to understand an absurdly complicated parking signpost on a random street in Williamsburg, Brooklyn, New York.

Context

  • Location: Williamsburg, Brooklyn
  • Subjects: Two young adult witches
  • Scene: Reading an overloaded street parking signpost
  • Props: Broom and rolled-up magic carpet
  • Style: Realistic candid street photography
  • Mood: Subtle visual comedy grounded in reality

Instructions

Place the two witches closest to the camera with their backs slightly turned, heads tilted upward as they carefully scrutinize a tall signpost on the right side of the street. One has ash-balayage hair and holds a broom; the other has long wavy auburn hair and carries a rolled-up magic carpet. Cover the pole with numerous unique, realistic NYC-style regulatory signs for parking restrictions, street sweeping, permits, towing, loading zones, and vehicle classifications without repeating signs. Among them, naturally include official-looking signs reading “NO BROOM PARKING • ZONE C”, “MAGIC CARPET LOADING ZONE • 15 MIN LIMIT”, and “REINDEER PARKING • PERMIT REQUIRED DEC 24–25 • VIOLATORS SUBJECT TO NAUGHTY LIST.” Make these fantasy regulations visually indistinguishable from legitimate municipal signage.

🤳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!