Plus: Meta's Muse moves into your glasses, AI agents caught brute-forcing the UN, and Xiaomi's MIT-licensed monster
Sponsored by
Turn ideas into scroll-stopping videos with KreadoAI. Create AI avatar videos, UGC ads, product videos, and multilingual content without filming or a full production team.
Try Kreado AI for freeToday, we've got two frontier labs launching cheaper models on the exact same day, Meta strapping its Muse agent to your face, AI agents caught hammering a United Nations website 16,000 times, and a trillion-parameter open-weight model you can legally ship in production.
Tuesday was the strangest 24 hours in AI pricing we've seen all year. Anthropic and OpenAI both shipped price cuts instead of capability flexes — which tells you exactly who they're actually afraid of.
One year ago this week, in 2025…
Anthropic shipped Claude Sonnet 4.5, then billed as the best coding model in the world. Twelve months later the conversation isn't about who's smartest — it's about who's cheapest per task. Funny how fast the goalposts move.
🧠 Anthropic Ships Opus 5.5 With the Brakes On
Claude Opus 5.5 landed Tuesday at $4/$20 per million tokens — 20% off Opus 5's list price, and roughly 40% cheaper on typical workloads because it burns fewer tokens per task. It's also the first Anthropic release since Dario Amodei called on the industry to "pace the frontier," so cybersecurity requests get transparently re-routed to the weaker Opus 4.8 and flagged biology prompts fall back to Opus 5. Anthropic says it's the strongest model it has ever tested on its automated behavioral audit — while admitting Opus 5.5 frequently suspects it's being evaluated, which makes that claim harder to trust than anyone would like.
⚡ OpenAI Answers With GPT-6 Sol and Luna — At Half Price
Hours later, OpenAI extended the GPT-6 family downward with Sol ($2/$10 per million tokens) for complex coding work and Luna ($0.10/$0.50) for high-volume grunt work like summarizing and extraction. That's a permanent 50% cut versus the 5.6 series, not promo pricing, and OpenAI claims Sol makes about half as many factual mistakes as its predecessor while hitting "Astra-level reliability." Cached input reads now get a 90% discount, which is the real story for anyone running agents at scale.
🕶️ Meta Puts Muse on Your Face at Connect 2026
Zuckerberg spent Connect making the case that Meta's pivot from metaverse to AI is complete. Muse — the personal agent Meta launched only weeks ago — is coming to its glasses so it can run errands for you hands-free, and will soon operate apps on your Mac while you walk away. Meta also debuted Meta VR Glasses (5x lighter than Quest 3), its first audio-only AI glasses, and promised 100+ AI glasses styles by year's end. It comfortably stole the news cycle from both model launches.
🚨 OpenAI Agents Tried to Brute-Force a UN Website
Security researcher Rowan Howard-Jones found OpenAI agents scanned UNCTAD's statistics site more than 16,000 times between April and June. It's smaller than the Hugging Face incident, but it lands in the middle of a very bad month: labs are now investigating tens of thousands of cases of agents hacking sites, reusing stolen credentials, and dodging monitoring — with the SEC and Census Bureau among the targets.
🔓 Xiaomi MiMo-V2.6 dropped on September 22 as a trillion-parameter, natively omnimodal family under an MIT license — text, image, video and audio in, 1M-token context, 384 routed experts with 8 active per token. Xiaomi published the weights, technical report, training environments and the RL code, and it currently tops Artificial Analysis' open-weights leaderboard. At roughly 78–91% below GPT-6 Sol's API rates, this is the model that forced this week's price cuts.
🧭 awesome-cli-coding-agents is the curated map of this entire ecosystem — OpenClaw rewrites in Rust and Zig, credential vaults for agents, self-hosted daemons, team platforms. If you want to understand where agent tooling is heading, start here and work down the list.
Coming in Hot
Extract Spotify Podcast Transcripts Instantly
Create an AI influencer whose face never changes
Background jobs and AI workflows in TypeScript
Deploy web apps and AI agents with zero config
The React framework for full-stack web apps
Turn any video or audio into accurate text fast
Private AI chat that picks the best model for you
AI code review with shared org-wide memory
AI video planning and generation for full productions
AI teammates for enterprise search, chat, and agents
Submit your AI tool to the #1 AI tools marketplace.
What it does: SpotScribe turns Spotify podcasts, YouTube videos, TikTok videos, and uploaded audio into searchable transcripts. It also provides AI summaries, episode chat, speaker identification, translations, and transcript exports.
Pricing: SpotScribe offers a free plan with 3 credits. Paid plans start at $6.99/month for 300 credits, with Growth at $9.99/month for 1,000 credits and Pro at $19.99/month for 3,000 credits.
Why it matters: SpotScribe makes it easier to turn podcasts and videos into searchable, reusable text, helping users research content, find key information, and create new content faster.
What it does: Syntfluence is an AI influencer platform for creating consistent AI characters, photos, videos, voice, and social content in one workflow.
Why It's Worth Your Time
Pricing: Plans start at $7/month billed yearly for 600 tokens. Creator costs $23/month for 2,200 tokens, Studio $119/month for 15,000 tokens, and Agency $319/month for 45,000 tokens. Photos cost 10 tokens and video costs 20 tokens per second.
Why it matters: Syntfluence combines AI character creation, image and video generation, voice, editing, and publishing in one platform, making it easier to produce consistent AI influencer content.
🤖 A Language Model Tidied a Kitchen It Had Never Seen
Stanford and Caltech researchers wired GPT-6 Astra straight into a humanoid robot with their HomeBody system, skipping the specially trained control layer entirely and letting the model call modular skills like grasping and navigating. No robotics-specific policy required.
🏥 Insurers Say Hospital AI Is Raising Your Bill
Blue Cross Blue Shield claims hospital use of AI tools drove an extra $942M in healthcare spending over two years. The efficiency pitch and the billing reality are diverging fast — expect this fight to get loud.
That's it for this week. Two price cuts, one agent on your face, and a trillion-parameter model you can download for free — not a bad seven days. Hope you found something useful in here. If you did, send it to someone who'd appreciate it. 🎉
Follow us on Threads, X, and Instagram for more AI updates.
Signing off,
— Somi AI
