Google

Gemini models and the Gemini API, Gemma open models, and Google’s developer tools for agents. Every brief item about Google, newest first, each with its primary source.

Subscribe via RSS Updated · Oct 14, 2026

Where it stands

Google has shipped on two tracks this month. On the open side it released EmbeddingGemma 2 on October 6 under Apache 2.0, then open-sourced the AQuA production-agent auditor and the ML Drift on-device GPU inference engine on October 8. On October 13 it released Gemini 4 Argon, its new frontier model, at an introductory $2/$10 per million tokens until November 30 and $4/$20 after, and GitHub made it generally available in Copilot the same day.

The trend is that Google is competing on price and distribution at once: a frontier model launched at half its list price, available on day one in Copilot as well as its own Antigravity, AI Studio and Gemini app, with open models and tooling keeping developers in its ecosystem below the frontier.

What to watch next: whether the introductory price is extended past November 30, how Argon’s 1M unified token limit and continuation protocol behave in long agent runs, and what Argon costs per finished task compared with other frontier models.

My advice: use the next seven weeks to run Argon on your own evals while it is cheap, but make the decision at list price; and if you use Gemma or EmbeddingGemma locally, look at the vLLM changes Google upstreamed, since they are free speed.

Darius · Updated Oct 14, 2026

Timeline 2 items

  1. Oct 14 · Wed · 2 items