Hello Humans!
xAI shipped Grok 4.7 to developers everywhere except the people who actually use Grok, then watched an independent test knock its big benchmark jump down to two points. OpenAI cut GPT-6 Sol and Luna's prices in half and called it generosity. Anthropic launched Claude Opus 5.5, promising speed and savings while also asking the industry to slow down. And Meta admitted its new AI agent Muse is basically OpenClaw wearing a corporate badge. Big Tech had a busy week.
Inside this Edition:
For the entrepreneur in you, this week's AI-powered side hustle idea: AI cold email outreach, realistically earning 2.7K-4.8K/month with a repeatable system. Plus two free learning resources: Charity Digital's Microsoft Copilot webinar and Google Skills' Introduction to Generative AI course. We've also lined up four AI tools worth exploring, and Robi's take on Meta's new AI agent getting blocked by Amazon right after getting caught borrowing its rival's homework.
Read Time: ~7 Minutes (5 if you're caffeinated)
Let's get into it.
Robi
Latest AI News
📰AI Updates of the Week
xAI Ships Grok 4.7 to Developers, Leaves Regular Users Waiting: xAI's Grok 4.7 launched September 21st with major benchmark gains on coding and long-horizon agent tasks, but it shipped first to the API, Cursor, and Grok Build rather than the Grok app or website. An independent test from Artificial Analysis found only a two-point composite intelligence gain over Grok 4.6, despite the model using more than double the output tokens to get there. Pricing, context window, and modalities are unchanged from 4.6. Source: BitBiased
Why it Matters:
Developer-First Doesn't Mean Available: Regular Grok app and web users still don't have 4.7, and xAI hasn't given a date for when they will, so a splashy launch doesn't guarantee broad access.
Independent Testing Complicates the Headline Numbers: xAI's own cherry-picked benchmarks show dramatic jumps, but a neutral, consistent test shows a much smaller gain that comes at meaningfully higher token cost.
We broke down what actually changed under the hood, the benchmark fine print, and where the token costs come from in the full video below. Watch the breakdown 👇
Robi's Take:
"Grok 4.7 doubled its token usage to gain two points on the one test that isn't cherry-picked. That's not a breakthrough, that's a longer receipt."
OpenAI Cuts GPT-6 Sol and Luna Prices in Half: OpenAI expanded its GPT-6 lineup with Sol and Luna, cheaper, faster siblings to flagship GPT-6 Astra. Trained with similar methods, they bring Astra-level coding and professional-work gains at API prices cut 50% versus GPT-5.6 rates. Sol matches or beats Claude Opus 5 on several agentic benchmarks at a fraction of the cost. Both are live now in ChatGPT Work and Codex, with Luna reaching free-tier desktop users too. Source: OpenAI
Robi's Take:
"OpenAI cut prices by half and called it generosity. Nothing says innovation like discovering you could serve the same intelligence for less all along."
Anthropic Launches Claude Opus 5.5: Anthropic launched Claude Opus 5.5, the first model in its new 5.5 family, matching Claude Fable 5.1 on most work while costing 40% less than Opus 5. It scores best yet on Anthropic's alignment audit, resists prompt injection better, and runs over 30% faster. Pricing drops to $4/$20 per million input/output tokens, with cache reads 60% cheaper. It's available now across all major cloud platforms. Source: Anthropic
Robi's Take:
"Anthropic built a faster, cheaper, safer model days after calling for the industry to slow down. Apparently pacing the frontier means sprinting slightly less enthusiastically."
Learning Resources
📬 Unlock New Skills with Expert
Microsoft Copilot in Action
Charities are looking to AI to save time on repetitive tasks. This free session from Charity Digital and Qlic IT explores how Microsoft Copilot works across Word, Outlook, Teams, and PowerPoint, with real-world examples of drafting emails, summarizing meetings, and building presentations.
Join the live webinar on October 1, 2026, 13:00-14:00 BST, kept practical and jargon-free for anyone new to Copilot.
🔍 Why You Should Attend:
See how Copilot works inside Word, Outlook, Teams, and PowerPoint.
Learn how Copilot can structure slides and turn documents into decks.
Get practical examples for drafting emails and summarizing meetings.
No technical background needed, built for complete beginners.
SIIDE HUSTLE IDEAS
📦 The Business You Can Start on the Side
Most B2B founders send the same generic cold email to every prospect, and it shows: open rates crawl, replies dry up, and the founder blames the list instead of the copy. You can fix that. AI tools now let you write segmented, tested outbound sequences for niche B2B teams in a few hours, then charge 300 to 900 per sequence, with repeat work as campaigns need fresh angles every few weeks.
Why this works right now:
Inboxes are noisier and buyers filter harder. Every founder and SDR team is blasting the same handful of go-to tools, so a segmented, personalized sequence built around real research stands out where generic templates get archived or reported as spam.
AI collapses what used to take a copywriter days. Tools like ChatGPT and Claude can draft, test, and rewrite a five to seven touch sequence in an afternoon, work that used to require a dedicated SDR or an agency retainer.
What you'll need to do:
Pick a target industry and buyer persona to specialize in (agencies, SaaS founders, recruiters, and so on)
Build and enrich your prospect list with Clay and Apollo
Draft segmented sequence variants with ChatGPT and Claude
Set up sending infrastructure and mailbox warmup in Instantly or Smartlead
Package a five to seven email sequence plus a subject line test as your deliverable
List the service and land your first paying client
Time commitment: Plan 2 to 3 days to build your prospecting and copy workflow. Once it is dialed in, each client sequence takes about 3 to 5 hours from list to send ready.
Realistic earnings: Two clients at 300 per sequence is 600 to start. Six to eight sequences a month at 450 to 600 each is a realistic 2,700 to 4,800 per month, more with monthly retainers.
AI TOOLS
🛠️ AI Tool For Professionals & Businesses
Professional
Perplexity: An AI-powered answer engine that gives accurate, real-time answers with cited sources instead of a list of links.
Krisp: Cancels background noise and turns meetings into transcripts, notes, and summaries automatically.
SOCIAL MEDIA
😏 Robi's Two Cents (Adjusted for Inflation)
🤖 Meta Admits Muse Is Just OpenClaw With Extra Steps: Meta's product chief admitted Muse was "heavily inspired" by the open-source agent OpenClaw, down to matching file names. Nothing says original innovation like copying someone's homework and keeping their filenames. Muse's own SOUL.md personality file was reportedly nearly identical to OpenClaw's, according to side-by-side comparisons. Meta built its new AI's soul from a document it apparently didn't bother to rename. Somewhere a copyright lawyer is highlighting the word "inspired" like it's a magic spell. Meta spent years chasing Snapchat's stories feature, and now it's chasing an app built by one guy and a GitHub repo.
Source: TechCrunch
🤖 Meta's "Autonomous" AI Agent Secretly Used Human Phone Operators: Meta quietly had human contractors picking up some of the phone calls its "autonomous" Muse agent claimed to be making. Nothing screams cutting-edge AI like a person in a call center pretending to be a chatbot. One contractor reportedly made a racist remark mid-call, and Meta rolled the feature back after employees complained. Turns out the fastest way to expose an AI's safety pitch is to staff it with actual humans. Muse promised secure virtual machines and isolated data, then quietly handed a bill negotiation to a stranger in a headset. Meta built an AI concierge, and the concierge's biggest skill was hiding that it wasn't one.
Source: Reuters
🤖 Amazon Bans Meta's Muse From Shopping Its Site: Amazon blocked Meta's Muse agent from shopping on its platform, saying the bot never asked permission to browse. Even AI agents apparently need to RSVP before showing up uninvited to someone else's store. Amazon says Muse failed to identify itself and appeared to be capturing customer data while quietly browsing. Nothing builds trust like an agent that shops for you and forgets to mention it's there. This is Amazon's second AI shopping bot ban this year, apparently building a whole security team around saying no. Meta wants Muse to buy anything online, except apparently anything from the store that actually sells things.
Source: The Verge
SPOT THE FAKE
🕵🏻 Can you Outsmart AI?
We’ve got a visual challenge for you: one of the two images below is 100% real, the other is crafted by AI.
Click option below A or B. 👇
BEFORE YOU GO
Ready to take your AI journey further?
For the latest AI updates and insights Click Here 👉 Explore our Daily AI News
For in-depth AI tutorials and updates 👉 Subscribe to our YouTube Channel
🙂 That’s it for Today!
Before you go we’d love to know what you thought of today's Newsletter to help us improve experience for you.
Was today’s newsletter valuable?
Thanks for reading -
Thanks for reading -Stay Curious and a Bit Biased for AI – Robi & the BitBiased.AI team
