Big week in AI and it's only Thursday. Anthropic gave away their best cheap model, Google cut the price of AI image generation to almost nothing, and the UN is fixing to hold a meeting about rules for all of it. Let's get into it.
Claude Sonnet 5 Is Free Now And It's Legit
Anthropic rolled out Claude Sonnet 5 on June 30 and by July 1 it was already the default model for every Free and Pro user. Not a watered down version either. They're calling it the most agentic Sonnet they've ever shipped, and by their own benchmarks it's running right up against Opus 4.8 on a lot of tasks.
This matters because "free tier" usually means "the model we don't care about." Anthropic just handed regular users a model that's competitive with their flagship. That's a real shift in how these companies think about who gets the good stuff.
My take: this is Anthropic playing offense, not defense. Google and OpenAI have been trading punches on price and speed all year, and giving away a near-flagship model for free is how you keep people from wandering off to try the other guy's app. Good for us regular users though. I'll take a free upgrade any day of the week.
Google Just Made AI Images Dirt Cheap
Google dropped two new image models the same day, June 30. Gemini 3.1 Flash Image runs fifty cents per million input tokens and three bucks for output. Gemini 3 Pro Image is pricier at two dollars in and twelve out, but that's still cheap for what it does.
Why it matters: image generation used to be the expensive, slow part of building an AI product. When the cost drops this much this fast, every app with a "generate an image" button gets cheaper to run overnight. That trickles down to whatever you're using day to day, even if you never touch Google's tools directly.
Robert's take: price wars are good when you're the customer and bad when you're trying to build a moat around image gen as a business. If your whole pitch was "we make AI images," you'd better have something else going for you now, because the underlying cost just went out the floor.
The UN Wants In On AI Rules
The UN Global Dialogue on AI Governance kicks off in Geneva on July 6. Member states are getting together to hash out how to manage this stuff internationally. A UN panel already put out a warning that the rules can't keep up with how fast the capabilities are moving.
This one matters because right now every country is writing its own AI playbook and none of them agree. That's fine for now, but it gets messy fast once these models are running hospitals, courts, and power grids in different countries under different rules.
My honest take: I don't expect Geneva to produce anything with teeth this early. These things move slow and the tech moves fast, that's just the nature of it. But somebody's got to start the conversation, and better it starts now than after something breaks bad enough to force the issue.
That's the roundup. A smarter free model, cheaper images, and the grown-ups starting to talk about guardrails. See y'all tomorrow.