It's Monday, August 24th: An anonymous model called Ox Alpha beat GPT-5.6 on coding in one independent test and nobody will say who built it, and CBRE's annual count puts New York ahead of the San Francisco Bay Area on tech headcount for the first time in thirteen years.
Covering what’s happening on the ground in AI, every Monday.
1️⃣ STEALTH MODEL: A Free Coding Model Beat GPT-5.6 In One Test And Nobody Will Claim It
Image from Open Router
A reasoning model called Ox Alpha appeared on OpenRouter on August 20 with a million-token context window, free access and no named maker, and within days it was posting coding scores people are still arguing about.
Ox Alpha lists on OpenRouter under the provider name Stealth, with a 1,048,576-token context, up to 131,072 completion tokens, text, image and video input, and $0 for prompt tokens, completion tokens and cache reads. OpenRouter routes the traffic and states plainly that it is not the developer, owner or provider.
The 80% that traveled around the timeline came from one independent run over 10 hand-picked coding tasks, where Claude Fable 5 scored 65% and GPT-5.6 Sol scored 52%. Community measurement across the full DeepSWE set puts Ox Alpha closer to 63%, and the model has not entered an official leaderboard.
Independent researcher Ben Davis matched Ox Alpha's token counts exactly against Z.ai's GLM-5.3 across 25 prompts, with a constant 75-token gap he reads as a hidden wrapper. The community tool modelprint hit on six of nine tokenizer probes, video handling matched GLM-5V-Turbo across four samples, and the same API restrictions showed up on Z.ai's own GLM-5.3 listing two days before Ox Alpha went live.
OpenRouter's own listing says prompts and completions for this model are retained by the provider and are not used for training, and that provider has no name. It sits on a platform Stripe agreed to buy for more than $7 billion, reported four days before Ox Alpha appeared.
If Ox Alpha is already in your model picker, you are routing production code to a provider nobody has named, and that provider keeps your prompts and completions. Run it on throwaway work if the free tier is worth it to you, keep client and proprietary code off it until somebody claims the model, and plan for the price to arrive the day the preview ends.
2️⃣ TALENT CROWN: New York Has More Tech Workers Than The Bay Area, And Far Fewer AI Ones
Image from CBRE
CBRE's annual tech talent count put New York Metro ahead of the San Francisco Bay Area on headcount for the first time in the survey's 13 years, 394,300 workers to 375,730, while the Bay Area held a wide lead on AI talent.
New York added 30,640 tech workers between 2022 and 2025, up 8.4%, while the Bay Area shed 23,900, down 6%. CBRE credits New York financial-services firms hiring technical and AI staff, plus AI startups taking space in Midtown South. Toronto added 75,000 jobs over the same three years, the largest gain on the continent.
Headcount is one metric out of 13. On CBRE's full scorecard the Bay Area is still first, then Seattle, then Toronto, with New York in fourth.
The AI gap is not close. The Bay Area holds 98,699 AI-skilled workers against New York's 67,949, and it has taken 80% of all US AI venture funding since 2020. AI roles are now 57% of Bay Area tech postings, up from 20% in mid-2022, against 31% nationally.
The AI-skilled workforce across the US and Canada reached 751,000 by June 2026, up 45% in a year. Colin Yasukochi, who runs CBRE's Tech Insights Center, expects the Bay Area to stay the industry's center, and says that as it grows, "that spreads out to all the key markets."
If you are hiring AI engineers, these numbers describe a split rather than a handoff. New York is where the volume moved, and the Bay Area is still where the specialists are. Price New York offers against finance rather than against startups, and assume anyone you want in the Bay Area has a market where 57% of open tech roles are bidding for the same skills.
📰 Other Headlines
TEEN MODE: OpenAI rolled out ChatGPT for Teens for 13 to 17 year olds, with Study Mode, Quiet Hours, parental notifications and homework reminders that fire when it detects cheating.
SOL GETS CHEAPER: OpenAI cut GPT-5.6 Sol API pricing by more than 20% for three months, to $4 per million input tokens and $20 per million output, down from $5 and $30.
MADE WITH AI: Apple Music will tag tracks built with a material amount of AI later this year, with labels and distributors responsible for declaring it, after Apple said more than a third of monthly uploads are fully AI-generated while AI music stays under 0.5% of listening.
BILLION DOWNLOADS: Google's Gemma open models passed one billion cumulative downloads with more than 100,000 community variants published in two years, including builds running onboard satellites.
CLAUDE GOES TO SCHOOL: Anthropic opened Claude Academy, a hub of courses, tutorials and completion badges built around the 4D AI fluency framework it uses to onboard its own staff.
TWENTY-FIVE REGIONS: AWS turned on cross-region inference for GPT-5.6 Sol, Terra and Luna on Amazon Bedrock, routing requests across more than 25 regions, with a US-only profile for workloads that have to keep inference inside the country.
KOREAN INFERENCE: Jensen Huang met Rebellions CEO Sunghyun Park in Santa Clara to discuss a partnership, an investment or an outright acquisition of the Korean inference chip designer, which has raised about $850 million from SK Hynix, Samsung Ventures and Arm.
AGENT INDEX: Firecrawl shipped Developer Index, a search layer over 70 million-plus READMEs, docs, issues, pull requests and OpenAPI specs, scoring 0.63 recall@10 on its open DevDex benchmark of 1,179 real developer queries.
🎓AIC AUTOMATIONS
It Runs While You Sleep.
Give an AI a job and a time, in plain English, and wake up to it done. You do not go hunting for a settings screen. You open the tool and ask it.
Image from Claude
How to set one up
Open Claude and go to Scheduled tasks.
Click New task, then Create with Claude.
Claude drops a prompt into the chat for you. It reads: "I want to set up a scheduled task. Briefly explain how scheduled tasks work in Cowork, then ask me a few questions to figure out what I'd like Claude to do and when it should run."
Send it. Claude explains how scheduled tasks work, then asks you what you want done and when.
Answer its questions. It builds the task and shows you the schedule.
(Codex works the same way: click into scheduled tasks and it opens a session with the prompt already written for you.)
You never write the instructions. You answer a short interview. Scheduled tasks sit on the paid plans, so check yours first.
Six worth stealing. The first four only read and report, so they cannot break anything. The last two write you a draft and stop.
Tomorrow’s info, tonight. Every evening, a rundown of tomorrow's meetings: who each person is, and what you agreed with them last time.
One narrow beat. Every morning, anything new on one specific thing you actually need to track. Narrow beats broad. "AI news" is useless, "our competitor's pricing page" is not.
Watch this page. Tell me when a particular page changes. A policy, a job board, a price, a filing.
The Weekly close. Friday afternoon, what happened this week, what slipped, and what carries into Monday.
What needs a reply. Each morning, the handful of emails that actually need you, each with a reply already drafted. Nothing sends.
Notes to actions. After your weekly team meeting, turn the notes into owners, deadlines, and a follow-up message ready to go out.
Pro tip: start with one of the first four. Something that only reads and reports can be wrong and you will just notice and fix it. Something that sends or posts can be wrong once and you spend the morning undoing it. Then check on it after a week, because a task you stop opening can quietly pause itself.
🫵 Want your message in front of 200,000 AI builders?
Our partners and sponsors get exclusive placements across the newsletter and access to AIC's in-person network — demo nights, dinners, hackathons, and forums across 180+ chapters.
For all inquiries, send us a note at [email protected].
The AI Collective is built by volunteers across 180+ chapters in 40 countries.
Thank you to the thousands of volunteers around the world who make this work possible. We truly could not do this without you.
🧑💻 About the Editors
About Noah Frank
Noah is a researcher, innovation strategist, and ex-founder thinking and writing about the future of AI and the workforce. His work and body of research explores the economics of emerging technology and organizational strategy. Outside of AIC, Noah heads research for Centaurian AI.
About Lindsay Gross
Lindsay is an AI engineer, researcher, and writer focused on how AI systems behave in practice and what it takes to make them safe. Her work sits at the intersection of AI safety, governance, and product design, and at AIC she writes about the questions that matter most as these systems scale.