Google ships Gemini 4 Argon, and ColGrep-Agent moves coding-agent search onto your laptop

Gemini 4 Argon (gemini-4-argon) is the first model in Google's Gemini 4 generation. Google built it for long, multi-step work in coding, finance, law, security and science. It handles up to 1M tokens of input and output combined, and the API gets a new continuation protocol for stitching long tasks together. Developers can use it on the paid Gemini API tier and in AI Studio. Ultra subscribers also get it in Antigravity, Google's coding-agent tool, and Pro and Ultra subscribers get it in the Gemini app and AI Mode in Search.

At $2 per million input tokens and $10 per million output (cached input is 95% off), it's cheap for a top-tier model. Google claims first place on the Vals Index of professional tasksVendor claim and says indirect prompt injection, where hidden text in a web page or email hijacks an agent, succeeds only 0.7% of the timeVendor claim. Artificial Analysis has it at 53, 8th of 216Independent, so it doesn't lead overall. It also showed up in GitHub Copilot the same day, which makes it easy to test against whatever you use now.

Before a coding agent can fix a bug, it spends turns working out where to look: grep, open a file, follow a reference, repeat. ColGrep-Agent is a 2-billion-parameter search agent that does that part on your machine. You ask it a question; it runs LightOn's semantic code search ColGrep, reads the likely files, follows references, and hands back file paths, line ranges and the code itself. Install it, then run the one line that matches your coding agent:

brew install lightonai/tap/colgrep
colgrep --install-claude-code   # Claude Code
colgrep --install-codex         # Codex
colgrep --install-opencode      # OpenCode

You don't need an API key or a separate inference server.

LightOn plugged it into Claude Code and ran 52 questions across 13 repos the model hadn't seen. Compared with plain grep, API tokens dropped 34% and cache-aware API cost dropped 40%, with no statistically significant change in accuracyVendor claim. If agent bills hurt, it's a cheap experiment: let the frontier model do the reasoning and the small local model do the digging.

You can now install Android apps you vibe-code in Google AI Studio onto your own device through Google Play. A regular Google account is enough; you don't need a developer account. The apps stay out of public search and listings, and only their creator can see them.

Until now, getting a homemade app onto your phone meant publishing it to everyone or sideloading an APK, so a lot of people stuck with web apps. Now you can try an internal tool or a personal automation as a native app without that detour. If you decide to go public later, you move to a developer account and go through the normal review.