Sonnet 4.6, Grok’s 4 Agents, and Meta’s Manus in Ads Manager | Weekly Digest
PLUS HOT AI Tools & Tutorials
Hey! Welcome to the latest Creators’ AI Edition.
This week was genuinely one of the wildest in the AI race so far.
Three major model drops in a single week (Sonnet 4.6, Grok 4.20, Gemini 3.1 Pro), Meta quietly slipping an AI agent into every advertiser’s dashboard, and the India AI Summit giving us the most meme-worthy moment of the year — all in seven days. 🤯
If you blinked, you missed at least two product launches. Let’s catch you up.
Featured Materials 🎟️
News of the week 🌍
Useful tools ⚒️
Weekly Guides 📕
AI Meme of the Week 🤡
AI Tweet of the Week 🐦
(Bonus) Materials 🎁
Keep your mailbox updated with practical knowledge & key news from the AI industry!
Featured Materials 🎟️
Claude Sonnet 4.6: Opus-Level Intelligence at 1/5 the Price 🧠
Anthropic released Claude Sonnet 4.6 on February 17 — just 12 days after Opus 4.6 — and it’s being called a “seismic repricing event” for the AI industry. Performance that previously required an Opus-class model is now available at Sonnet pricing: $3/$15 per million tokens (vs. Opus at $15/$75).
What’s new?
Computer use jumped to 72.5% on OSWorld — nearly a fivefold improvement from Sonnet 3.5’s 14.9% just 16 months ago. This is the capability that unlocks automation of legacy software without APIs.
1M token context window (in beta) — enough to hold entire codebases, lengthy contracts, or dozens of research papers in a single request. Double the largest window previously available for Sonnet.
60.4% on ARC-AGI-2 — above most comparable models, though still behind Opus 4.6 and Gemini 3 Deep Think.
Box tested it extensively: 77% accuracy on heavy reasoning tasks (up from 62%), 94% on retail tasks, 88% on public sector work, 78% on healthcare.
GitHub’s VP of Product Joe Binder confirmed: “Already excelling at complex code fixes, especially when searching across large codebases.”
Sonnet 4.6 is now the default model for Free and Pro plan users on claude.ai and Claude Cowork. Available on all Claude plans, Claude Code, the API, and all major cloud platforms.
Is it a breakthrough?
This isn’t just a model update — it’s a business model disruption. When Opus-class intelligence becomes available for a few dollars per million tokens, companies that were cautiously piloting AI agents suddenly face a fundamentally different cost calculus. The agents that were too expensive to run continuously in January are suddenly affordable in February. VentureBeat called it “the headline that matters most.”
Source: CNBC | VentureBeat | TechCrunch
Grok 4.20: xAI Didn’t Release a Model — It Released a Team
Elon Musk didn’t announce a launch window for Grok 4.20 — he just posted that it was live on February 17. And it’s the most structurally different Grok release since the original.
Instead of a single model answering your question, four specialized AI agents now work together on every query:
Grok (Captain) — coordinates, synthesizes, delivers the final answer
Harper — research, fact-checking, real-time data from X’s firehose (~68M English posts/day)
Benjamin — math, code, logic, step-by-step reasoning
Lucas — creative thinking, divergent ideas, balance
They run in parallel, debate each other’s outputs, and peer-review before you see a single word.
What’s new?
Hallucinations dropped 47–65% according to early users. The peer-review mechanism catches contradictions before output.
“Rapid learning” architecture — the model improves weekly based on user feedback, with release notes published alongside every update. Musk: “Unlike prior versions of Grok, 4.2 is able to learn rapidly.”
Grok 4.20 Heavy (SuperGrok Heavy tier) scales up to 16 agents for extreme-complexity tasks.
In Alpha Arena Season 1.5 (live stock trading), Grok 4.20 variants were the only profitable models — turning $10K into $11K–$13.5K while OpenAI and Google rivals finished in the red.
Medical document analysis: Upload lab reports, imaging results, prescriptions for AI-generated analysis.
Available to SuperGrok (~$30/month) and X Premium+ subscribers. Free users can also access the multi-agent system on grok.x.ai.
This is the first consumer-facing AI from a major lab where multiple named agents visibly reason together. Love it or hate it, it’s a genuinely different approach.
Source: NextBigFuture | AdwaitX | EONMSK
Meta Puts Manus AI Inside Ads Manager — Every Advertiser Now Has an AI Agent 📊
About a month after acquiring Manus for $2 billion, Meta has embedded the autonomous AI agent directly into Ads Manager. This isn’t a chatbot that writes copy when asked — Manus acts as an autonomous operator inside campaigns.
What it does:
Automates report building and campaign analysis
Conducts audience research without switching platforms
Generates insights from performance data
Handles recurring tasks on a schedule
All Meta advertisers can access Manus via the “Tools” menu in Ads Manager right now. Some users are also getting pop-up prompts encouraging activation.
Should you care?
Yes — but temper expectations. Jon Loomer noted that the current integration is more of a redirect to Manus AI’s LLM interface than a seamless in-platform experience. The real promise is ahead: as Meta tightens the integration, expect autonomous campaign optimization, A/B testing, and creative iteration without human prompts. Alexander Zakharov warned: “Brands should expect to lose visibility into campaign decisions. Define your guardrails now.”
Source: Social Media Today | Search Engine Land | Jon Loomer
News of the week 🌍
Gemini 3.1 Pro Drops (Feb 19) 🧪 — Google released the first “.1” increment in Gemini history — skipping the usual mid-year “.5” cycle. Gemini 3.1 Pro scored 77.1% on ARC-AGI-2 (more than double Gemini 3 Pro’s 31.1%), hit 44.4% on Humanity’s Last Exam, and topped the APEX-Agents leaderboard for professional task performance. Priced the same as 3 Pro ($2/$12 per million tokens) — less than half the cost of Claude Opus 4.6. Available in the Gemini app, NotebookLM, AI Studio, and Vertex AI. Source
India AI Impact Summit 🇮🇳 — The biggest global AI summit of the year brought Altman, Amodei, Pichai, and Hassabis to New Delhi. India projected $200B+ in AI investment over the next 2 years and committed 20,000 GPUs for domestic infrastructure. Adani committed $110B for green AI data centers. Reliance Jio announced $109.7B for AI infrastructure. Anthropic opened its Bengaluru office and partnered with Infosys. OpenAI opened two new India offices and partnered with TCS. Source
Meta x Nvidia Multi-Year Deal 🤝 — Meta and Nvidia announced a massive multi-year partnership for Meta’s AI data center buildout. Includes Nvidia’s Vera Rubin platform, Spectrum-X networking, and Confidential Computing for WhatsApp. Meta becomes the first to deploy Nvidia’s Grace CPUs as standalone chips. With Meta’s $135B 2026 capex, the deal is “certainly in the tens of billions.” Source
Micron’s $200B Expansion 💾 — Micron plans a massive expansion to address what it calls the biggest memory chip shortage in 40 years. AI workloads are consuming memory faster than the industry can produce it. Source
Cohere Goes Multilingual & On-Device 🌍 — Cohere released a new family of open multilingual models supporting 70+ languages, small enough to run on everyday devices. The “small-but-capable” direction matters: companies want AI that’s cheaper, more private, and viable in regions where cloud costs are constraints. Source
1 in 3 Say AI Chatbots Are Their Worst Service Experience 😬 — New survey data shows nearly a third of customers rank AI chatbots as their worst customer service interaction. The trust gap should alarm every brand betting on AI-first support. Source
Google I/O 2026 Announced 📅 — Set for May 19-20 at Shoreline Amphitheatre. Google is making AI the headline from the start, with Gemini front and center alongside Android, Chrome, and Google Cloud updates. Source
Deutsche Bank Asked AI How It Plans to Destroy Jobs 🏦 — The bank literally asked an AI model to predict which jobs it would eliminate. The AI predicted handling 75% of all customer service interactions by 2026 and flagged media/entertainment as “likely to be disrupted.” But it also conceded “sectors of resilience” where deep empathy remains premium. Source
Useful tools ⚒️
⭐ Moda — (Our Top Pick) “Finally, AI designs you can actually edit.” Generate on-brand visuals and then tweak every element — fonts, colors, layouts — without starting from scratch. Launched Feb 18 on Product Hunt with 521+ upvotes. Great for marketers and social media managers who are tired of AI designs they can’t customize.
Origami.chat — Find your perfect leads with one prompt. Describe your ideal customer, and Origami pulls matching contacts with emails and context. 481 upvotes on PH (Feb 19). Sales teams, take note.
Reloop — Create winning ads without prompts or skills. Upload your product, and Reloop generates scroll-stopping creatives optimized for Meta and Google. Launched Feb 19 with 319 upvotes.
Monologue for iOS — Turn your voice into polished writing anywhere on iPhone. Think quick notes that come out as finished paragraphs. Launched Feb 19 — perfect companion for content creators on the go.
IMAI Studio — AI-powered design studio for enduring brand experiences. Trending on TAAFT this week (Feb 17) with 5,100+ views. Focuses on creating consistent, professional product design assets.
Share this post with friends, especially those interested in AI!
Weekly Guides 📕
Getting Started with Claude Sonnet 4.6 — Official model docs and migration guide from Sonnet 4.5
How Grok 4.20’s Multi-Agent System Works — Deep dive into Harper, Benjamin, Lucas, and the Captain
Gemini 3.1 Pro: Getting Started for Developers — Official guide with API access via AI Studio, Vertex AI, and Antigravity
Mastering n8n for AI Automations — Revisit our deep dive and hook up your workflows to the new models
AI Meme of the Week 🤡
The GPT-4o retirement hit the AI community hard. OpenAI removed it from ChatGPT, and the internet responded exactly how you’d expect:
Over 20,000 people signed a petition. A dedicated subreddit (r/4oforever) was created. People are literally mourning a language model. We’ve reached the stage where AI models get retirement parties and funeral memes. What a time to be alive. 😭
AI Tweet of the Week 🐦
The moment of the week happened at India’s AI Impact Summit when Prime Minister Modi asked all the tech leaders on stage to hold hands and raise them together in a show of unity. Everyone joined in — except Sam Altman and Dario Amodei, who were standing right next to each other and conspicuously raised their fists instead of holding hands. 🤜🤛
(Bonus) Materials 🎁
Anthropic & Infosys Partner for Telecom AI Agents — Claude heading into regulated industries at scale
Fresh Air: “Can the AI Chatbot Claude Be Contained?” — NPR’s deep dive into AI ethics, aired Feb 18
MIT: AI Models Perform Worse for Non-Native English Speakers — Important equity research on AI bias in language proficiency
MIT: Long Conversations Cause LLMs to Mirror User Viewpoints — Echo chamber effect in extended AI chats
Fortune: Thousands of CEOs Admit AI Had No Impact on Productivity — The “Solow Paradox” is back, 40 years later
If you missed our previous updates, don’t worry, here they are:
GLM-5, Gemini 3 Deep Think, and Anthropic’s $30B | Weekly Digest