Somebody just blew up the pricing logic for AI, and if you're paying for any AI tool or API right now, you need to know about it.

Chinese lab Moonshot AI released Kimi K3, and the headline number is almost hard to believe: 2.8 trillion parameters. For context, a "parameter" is roughly one tiny learned connection inside the model's brain. More isn't always better, but at this scale, it usually means the model can handle harder, more nuanced tasks without falling apart. K3 benchmarks within a few points of Claude's latest flagship and GPT-5 on general reasoning. On web development and frontend coding? It beats both.

Here's the part that matters to you: it's priced at $3 per million input tokens and $15 per million output tokens. That's the same as Claude Sonnet 5, which most people think of as a solid mid-tier workhorse, not a frontier model. The full model weights drop July 27, meaning technically-minded owners or the developers you hire could eventually run it themselves, on your own servers, with no per-token bill at all.

The practical upshot is this. If you've been paying premium prices for top-tier AI reasoning because "the cheaper models just can't handle it," that excuse is gone. Near-frontier capability at mid-tier cost is now a real thing, not a marketing claim.

You don't need to switch anything today. But if you're building automations, hiring a developer to work with AI APIs, or choosing between AI plans, the cost-to-capability math changed overnight.

·        Viktor: An AI employee that lives inside Slack and Teams, connects to 3,200+ tools, and fixes workflow problems before you notice them. For owners who want AI that acts, not just answers.

·        River: AI sales reps that join B2B calls instantly, run demos, handle objections, and close. For service or SaaS businesses losing deals to slow follow-up.

·        Timely: One keystroke pulls your calendar availability, auto-converts time zones, works across multiple execs, runs locally with no data sent to the cloud. For anyone who wastes 10 minutes copy-pasting "here are some times that work" emails.

·        Aye: A teachable browser intern for Mac and Windows that reads pages, plans steps, and handles clicking, typing, and research on its own. For owners who want to delegate repetitive web tasks without writing a line of code.

·        Paradigm: Turns any learning goal into a personalized, adaptive step-by-step path that adjusts as you go. For owners and team members upskilling in AI, marketing, or finance without time for formal courses.

·        The 'what to automate' framework every owner needs: Matt Wolfe tried to fully automate his YouTube thumbnail workflow and it failed. His three-bucket test: fully automate high-volume repetitive tasks, use AI to augment research and first drafts, and keep anything taste- or relationship-driven fully human. Free webinar walking through real examples, including the ones that flopped.

·        The AI divide is now inside companies, not just between them: Median AI spend is $11.38 per worker. Top-quartile firms spend multiples more and are compounding that advantage every month. If you haven't set even a modest per-seat AI budget, peers who have are already pulling ahead.

·        Canva Code 2.0 lets anyone build a website or app from a plain-language prompt: Describe what you want, get editable output, starting from templates, HTML, or existing Canva designs. Available now on free plans. If you're already in Canva for design, this cuts a separate web-builder subscription.

·        Anthropic's 4-stage framework for moving AI from one power user to a whole team: Break bottlenecks, add guardrails, measure real ROI, then automate maintenance. Worth reading if one person on your team is getting all the AI benefit while everyone else shrugs.

·        xAI open-sourced Grok Build — full coding agent, terminal UI, and plugin system free on GitHub: The complete source code is now public. Relevant if you're hiring developers who use AI coding agents or want to understand what these tools can actually do under the hood.

·        How the AI boom is creating a wave of new entrepreneurs: Forbes points out that the AI layoff coverage keeps missing the flip side: record new business applications and a specific set of skills separating the founders getting traction from the noise.

·        5 AI tools that make starting a business meaningfully faster: Inc. rounds up tools for validating ideas, building MVPs, and launching. LinkedIn data cited in the piece shows 85% of small businesses are already using AI for data entry, reports, and content.

·        Suno was hacked and its full training dataset was exposed: 113,000 hours of YouTube Music and Pond5 audio. If you're using AI-generated music in your marketing, this is a potential IP liability situation worth watching.

·        Social content is now tracked in Google Search Console: Instagram, TikTok, X, and YouTube posts now show up in Search Console data. If you run social alongside a website, you can finally see the search impact of your posts in one place.

·        Mira Murati's first open model is out: Thinking Machines released Inkling, a 975-billion-parameter model that reasons across text, image, and audio and can be fine-tuned for cost or performance. Built to be customizable rather than one-size-fits-all.

·        MarketBlazer launched an AI marketing platform for small businesses: Atlanta-based MarketBlazer announced a new AI platform aimed at small business marketing. Details in the press release are thin, but worth keeping an eye on as specifics emerge.

-Scott

SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.