The 'Free Players' Just Moved the Goalposts: Why AI Cost Is No Longer Your Excuse

There's a quiet shift happening in AI, and most businesses are looking the wrong way to see it.

Everyone's watching the premium models — the headline launches, the eye-watering benchmark scores. Meanwhile, the two biggest "free players" in the game — Meta and Google — have been busy making frontier-grade AI so cheap and so accessible that cost has basically stopped being a reason not to build.

Let me show you what I mean.

Meta: giving the engine away

Meta's Llama 4 family is open-weight. That means you can download the actual model and run it on your own hardware — no per-token bill, no vendor holding your data hostage. Llama 4 Scout runs on a single GPU, handles a 10-million-token context window (think entire document libraries in one prompt), and outperforms models it has no business beating at its size. Its bigger sibling, Maverick, goes toe-to-toe with GPT-4o-class models — at a fraction of the active compute, because it only switches on the parts of the model a given task needs.

Read that again: the engine room of serious AI is now free to download. For an Australian business worried about data sovereignty or vendor lock-in, that changes the whole conversation. You can run capable AI inside your own walls, on your own terms.

Google: frontier intelligence at vending-machine prices

On 19 May 2026, Google shipped Gemini 3.5 Flash. This isn't a budget model wearing a cheap badge — it beats Google's own flagship (3.1 Pro) on most coding and agentic tasks while running roughly 4x faster and costing about 40% less. It reads text, images, audio, video and PDFs, holds a million tokens of context, and lets you dial reasoning effort up or down depending on the job — minimal for speed, high when the problem is hard. On agentic tool-use benchmarks it's leading models that cost far more.

The price? Around $1.50 per million words of input. For most rea