Grok, ChatGPT, and Meta All Just Launched Shared AI Teammates You Can Give to Your Whole Team
AI has followed a pretty clear path. First it was a chatbot you typed questions into. Then it became a proactive personal assistant working in the background for you alone. Now it's becoming a shared teammate your entire team uses together, like a coworker with a desk. All three of the biggest players moved in the same direction within weeks of each other. xAI launched Team Bots, shared Grok agents your team configures around a role, loads up with files and app integrations, and invites into a Slack channel where everyone can work with it. OpenAI shipped its Dots agent with similar shared-channel capability. Meta launched Muse for Small Business. Same idea, three different platforms, all within a few weeks.
If you have even a small team helping with property management or your real estate business, this is the category to watch and probably the right time to start testing one. You can spin up an agent loaded with your lease terms, maintenance procedures, and vendor contacts, share it across your team in Slack or whatever you already use, and have it handle tenant questions, triage maintenance requests, and knock out admin tasks without anyone having to re-explain the context every single time. Grok Team Bots are in public beta on Teams and Enterprise plans right now, so this is real and available. Small operators stand to benefit from AI shifting to a shared team resource just as much as big companies do.
Read More →
Sources: x.ai, linkedin.com, mediarelease.co, agentriot.com, runtimewire.com, aishiftnews.ca, chasingnext.com, aiproviderindex.com, worldprogramming.org, pagegrok.org, usaaitools.com, writebotics.com
Google Just Launched Its Most Powerful AI Model to Take On OpenAI and Anthropic
"Frontier models" is the industry term for the most powerful AI systems any lab can build right now, the ones pushing the edge of what AI can actually do. Think Formula 1 cars. Not what most people drive, but where technology gets proven before it trickles down into everything else. OpenAI has GPT-6 Astra in that race, Anthropic has Claude Opus 5.5, and now Google has officially rejoined with Gemini 4 Argon.
Two things stand out. Argon outperformed both GPT-6 Astra and Claude Opus 5.5 on software engineering benchmarks, meaning it can work through longer, harder technical problems in one sitting than its rivals. It can also generate roughly 16 times more output in a single response than the previous Gemini generation, so it can hold a much longer train of thought and complete bigger jobs without stopping mid-task. Analysts say Google is back at the frontier but not clearly ahead, with Anthropic still holding a slim overall edge.
For most people, Argon isn't available yet. Access is starting with a narrow group of vetted cybersecurity organizations, and broader availability for paying users comes later with no firm date given. Pricing starts at $2 per million input tokens and $10 per million output tokens (tokens are the chunks of text an AI reads and writes, roughly three-quarters of a word each), roughly competitive with what OpenAI and Anthropic charge at the top tier. The frontier race is genuinely three-way now, and competition at that level historically pushes prices down and capability up across all the AI tools everyday users actually touch.
Read More →
Sources: cnbc.com, Manual, blog.google, 9to5google.com, alphaxiv.org, linkedin.com, the-decoder.com, siliconangle.com, futurumgroup.com, coursium.ai, apidog.com, aiintoai.vercel.app
A New Breed of Ultra-Fast, Ultra-Cheap AI Models Just Launched and They Only Answer Yes, No, or Pick One
A few weeks ago, a San Francisco startup called TypeSafe AI launched something called Jev and quietly started a new category in AI. Regular AI models, the kind behind ChatGPT, generate full sentences and paragraphs. They reason, they ramble, they take time, and they cost money per word. Decision models skip all of that. You send one a question with a list of possible answers, and it comes back with a choice, a score from one to five, or a yes/no probability. Jev costs less than five cents per million words of input, and output is free. Response times run as low as 70 milliseconds, roughly the blink of an eye. Within two weeks of Jev launching, Cloudflare released two of its own called Clef and Clef-flash, AWS released one called Strands Decider, and OpenAI announced a Decisions API at its developer event. Cloudflare's faster model, Clef-flash, comes back in under 40 milliseconds at the median.
You do not need to be a developer to understand why this matters. Think about every place a business needs a quick judgment call. Is this support ticket urgent, which team gets it, is this message spam, should this request be flagged or waved through. Normally you would either pay for a full AI model to reason through it (slow, expensive), or build a rigid rule system that breaks the moment something unexpected comes up. Decision models sit in between. Smart enough to handle new situations without retraining, but fast and cheap enough to run thousands of times a minute without the bill piling up. AI is quietly splitting into two lanes. Heavy models that think and write, and lightweight decision models that just route, filter, and triage at speed. Worth knowing the category exists even if you never touch the underlying tools yourself.
Read More →
Sources: aieatingtheworld.com, Manual, developers.cloudflare.com, devfeed.tech, systemonemodels.org, nvdium.com, arxiv.science, laurencemoroney.com, marktechpost.com, arxiv.org, the-decoder.com, developers.cloudflare.com
|
