Anthropic Quietly Ate OpenAI's Lunch
Reports this weekend put Anthropic's annualized revenue north of $30 billion. That is a wild number for a company that was bleeding cash not long ago.
OpenAI is sitting around $25 billion annualized and reportedly eyeing a public listing later this year.
Why it matters. The frontier model race is not a single horse anymore. Enterprise buyers are splitting their bets, and Claude is grabbing a big chunk of that spend. When the incumbent gets passed on revenue, the narrative shifts fast.
My take. I have been on the Claude bandwagon for a while, and this tracks with what I see in the wild. Developers and ops teams keep picking it for serious work. OpenAI still owns the brand, but brand does not pay the bills forever.
Google Is Going After the Cheap Seats
Google rolled out Gemini 3.1 Flash-Lite this week. 25 cents per million input tokens. 2.5x faster response times than the last Flash version.
That is not a flagship play. That is a volume play.
Why it matters. When a frontier lab ships a model this cheap and this fast, it squeezes the middle of the market. Smaller providers who were competing on price just got lapped. Teams running high volume workloads have a very good reason to switch.
My take. Cheap and fast beats smart and slow for 80 percent of real world jobs. Summarization, classification, routing, data extraction. None of that needs a $200 per million token model. Google knows this and is going after the volume while Anthropic and OpenAI fight over the top shelf.
AI Agents Are Scaring the Security Folks
The agent wave is here. Tools that browse, click, type, and buy stuff on your behalf. Cybersecurity folks are not sleeping well about it.
The concern is simple. You give an agent access to your email, your calendar, your browser. Now every phishing email and shady web page is a potential attack on your agent, not just on you. Prompt injection goes from lab curiosity to real threat.
Why it matters. Most companies are not ready. The security stack assumes humans make the click. Agents do not read warning banners. They just follow instructions.
My take. I am bullish on agents long term. But if you are deploying them in your business right now without scoped permissions and a human sign off on anything that writes or spends, you are asking for a bad quarter. Read only agents first. Write actions need guardrails. No exceptions.
Bottom Line
The money is moving. The prices are dropping. The risk surface is expanding. Pick the tool that fits the job, keep a short leash on anything that can act for you, and do not believe your own hype.