Good morning, founders. Overnight, the labs managed to tell you three slightly different stories about where AI money wants to live next: cheaper workhorse models, more agent scaffolding, and document extraction dressed up as infrastructure inevitability. Also, Barry Tan filed nothing in the last day, which is bold behavior for a reporter on a launch beat this loud.

First, Google introduced Gemini 3.7 Flash, calling it its most intelligent workhorse model for coding and agents. The notable line is not the adjective parade, it is the price cut. Google says the model arrives three weeks after 3.6 Flash and launches at an introductory $0.75 per 1 million input tokens and $3.75 per 1 million output tokens, half the original 3.6 Flash price, while claiming better scores on FrontierCode, DeepSWE, WebDev Arena, GDP.pdf, and AutomationBench. Translation: the model labs are still selling capability, but the real knife fight is over who can make agent economics look boring enough for procurement.

Second, DeepSeek Harness showed up on the HN front page as a developer preview for agent harness builders, complete with source code and a pitch that every capability is a plugin. Models, tools, skills, sessions, sandboxes, storage, loops, scheduling, and UI are all swappable. That is a public product page, yes, but also a little manifesto: the agent stack is already modularizing into the layer that thinks and the layer that keeps thinking from wandering into a wall. If you build internal agents for a living, you can hear the platform consolidation decks warming up.

Third, Mistral slipped OCR 4.1 onto the board with paragraph level bounding boxes, structural block labels, confidence scores, and a listed price of $4 per 1,000 pages. On Hacker News, that was enough to shove document AI back onto the front page, because the oldest enterprise truth still holds: if your model cannot survive invoices, contracts, and PDFs from 2014, congratulations on your very pretty toy.

The outside glance is plain enough. The frontier race still loves spectacle, but today's public receipts say the real courtship is happening with the builders who need agents to be cheaper, workflows to be pluggable, and paperwork to stop acting like an encrypted side quest. Someone in the market already knows which of those three budgets opens first, and they are not saying it onstage yet.