Google Releases Gemini 4 Argon to Developers and Subscribers

Google released Gemini 4 Argon, its new top model, on October 13, 2026. Developers can use it through the Gemini API, and Gemini Pro and Ultra subscribers can pick it in the Gemini app and in Search. Independent scores are mixed. Vals AI's overall ranking puts Argon first. Artificial Analysis, which runs the same tests across many models, gives it 53 on its overall score, behind Anthropic's Claude Opus 5.5 at 58. Google's own chart has Argon first on DeepSWE, a test of fixing real software bugs, at 77.9%. You don't need Google's app to try it. Argon is already an option in the coding tools Cursor, OpenCode and Devin. Google's announcement ↗

Anthropic Replaces Git After Merging 30,000 Code Changes Daily

Anthropic said on October 13, 2026 that it no longer uses git, the standard tool programmers use to track changes to code. Its engineers, working with Claude agents, now merge more than 30,000 code changes a day. Git-based tools could not keep up. So a team of seven people spent five months building a replacement called Scion. Anthropic says Scion uses about 200 times less computing power than its old git setup. The company is not releasing it publicly. The numbers give a concrete measure of how much code AI agents now write at one AI company. Anthropic's engineering post ↗

Fish Audio Releases Drama 3 Voice Model for Download

Fish Audio released Drama 3 on October 13, 2026. The text-to-speech model turns written text into spoken audio, and anyone can download it. You direct it in plain language. Change the tone partway through a sentence, or voice up to 16 characters in one scene. Businesses earning under $500,000 a year can use it commercially for free, and the license bans cloning a person's voice without their consent. Fish Audio's paid service charges $10 per million characters until November 12, roughly $5 for a full-length novel. Fish Audio's blog ↗

Liquid AI Releases LFM2.5 Model for AI Agents on Laptops

Liquid AI released LFM2.5-24B-A2B on October 13, 2026. The free-to-download model runs AI agents on your own computer instead of a company's servers. It switches on only a small part of itself for each word, which keeps it fast. Liquid says it produces 184 word pieces per second on Apple's M5 Max chip. In Liquid's own tests it beat Alibaba's Qwen model on agent tasks but scored lower on a standard bug-fixing test, 51% against 66.2%. Liquid uses its own license, so businesses should read the terms before using the model. Liquid AI's blog ↗

diagram-design Teaches Coding Assistants to Draw Clean Diagrams

diagram-design gained 9,867 stars on GitHub this week. The free instruction set teaches coding assistants such as Claude Code, Codex and Cursor to draw 44 kinds of diagrams, from flowcharts to timelines. Each diagram comes out as a single web page you can open in any browser or paste into a document. Read more ↗

cua Provides Open Tools for AI Agents That Operate Computers

cua is free, open-source software for running AI agents that use apps by clicking and typing. The agents work on Windows and Mac machines as well as Linux. It can run many of these computers at once. It also includes tests that measure how well an agent completes tasks on a screen. The project was on GitHub's monthly trending list. Read more ↗