Google opened Gemini 4 Argon to everyone on 13 October, and on Google's own benchmark table it leads Claude Opus 5.5 on automation work while Opus still wins in the terminal. These are Google's numbers, Argon first and Opus 5.5 second: - AutomationBench: 51.3% against 42.5% - DeepSWE v1.1: 77.9% against 74.2% - Terminal Bench 4.0: 57.4% against 66.4% - FrontierSWE v2: 55.0% against 62.3% It is in the Gemini API through AI Studio now, at an introductory $2 per million input tokens and $10 per million output, going to $4 and $20 after 30 November. The Gemini app has it for Pro and Ultra subscribers. The first hands on reports are more mixed than the table. ariwilson on Hacker News ran his own eleven bug fixes through it and it solved ten, but a median run cost him $0.52 against $0.18 for Sonnet 5.5 and took about four minutes against 32 seconds, because it wrote around 24k output tokens where Sonnet wrote about 3k. eyalitki gave it a payroll question, watched it get every edge case right, and then saw it work out 430 minus 60 as 410. Google's own engineers left it running overnight on ADK, their 200,000 line Python agent framework, and the changes it made are now live, with in memory session prep in long sessions down 40 percent. If you build with n8n or Claude Code, I would not move anything yet. An automation step that keeps failing on a long, many step task is the cheap thing to try it on at the introductory price, and for work in the terminal Opus 5.5 still has the better number. 🔗 https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/ The tester's numbers: https://news.ycombinator.com/item?id=50077034 Which step in your own workflows would you hand to a new model first?

I gave Claude Code one prompt at every effort level and asked each run to build a living New York City I could walk through in the browser. Extra high won. 🔗 Official links from the video: ➡️ https://go.genovaflow.com/skool-claude-code-at-every-effort-level-free-video-editing-watch Every run was Opus 5.5 in its own fresh session, one shot, scored against the same 20 checks. Low finished in 18 minutes for about $4 and scored 2 out of 10. Extra high used 502K tokens, fewer than high, cost $23 and passed all 20 checks. Ultracode took 3 hours 39 minutes and $148 to score 6, the same as high. 💡 What you'll learn ✅ What the effort setting actually changes in a build ✅ Why extra high passed every check on fewer tokens than high ✅ When max is worth $45 and when Ultracode is worth $148 📎 Both free files are attached to this post ➡️ new-york-city-prompt.zip The exact prompt every effort level got, word for word. Open a fresh Claude Code session in an empty folder, pick an effort level and paste it, and you are running the same test on your own plan. ➡️ claude-code-video-editing.zip The video editing skills I promised on screen. Drop your raw footage in a folder, open Claude Code there and ask for what you want, like "cut the ums and add captions" or "make three shorts from this". It transcribes the footage, cuts it on word boundaries, builds titles and captions with HyperFrames, exports with ffmpeg, and checks its own render before you see it. It always shows you the plan and waits for your yes before it cuts anything. ⚙️ Setting it up takes one prompt Unzip the folder, open a terminal inside it, run claude, and paste this: Set up the video editing skills in this folder for me. Read install.md and follow it from start to finish. First check my computer, then ask me one question: ElevenLabs (paid API) or Whisper (free, needs a GPU), and tell me which one you recommend for my machine. If I pick ElevenLabs, ask me for my API key. Do everything else yourself, test that it works, then tell me it is ready and wait for my footage.

My usage limit used to die inside the first hour on the 5x Max plan. Same work now finishes with the meter around 30 percent. 🔗 Official links from the video: ➡️ rtk: https://github.com/rtk-ai/rtk ➡️ caveman: https://github.com/JuliusBrussee/caveman ➡️ The CLAUDE.md from this video is attached to this post. I go through the eleven changes that took my Claude Code token usage down about 70 percent on the same work. Two of them are free repos you install once, rtk for what goes into the context window and caveman for what comes back out. The other nine are habits that are already sitting in your terminal and cost nothing. I show the before and after on my own usage screen, the exact numbers each change moved, and where I think the honest ceiling is on each one. 💡 What you'll learn ✅ How to install rtk so a 2000 token command dump comes back as 400 ✅ How to add caveman so Claude answers in facts instead of a wall of text ✅ How to read /context and find the MCP chunk that is eating your window ✅ How to run /compact yourself at 60 percent instead of letting it fire at 100 ✅ How to hand a long session over to a fresh one without losing your context ✅ How to keep the prompt cache alive by staying on one model per session ✅ How to keep CLAUDE.md under 200 lines and feed it markdown instead of PDF ✅ How to set /model opusplan so Opus writes the plan and Sonnet writes the code The attached CLAUDE.md has all eleven already written up as rules, including the always on rtk line and the handoff command. Drop it in the root of any project and Claude Code reads it on every turn. It is 158 lines, because rule eleven is keeping it under 200. If you only do one thing before watching, run /context and look at what is actually filling your window. Mine was an MCP block from servers I connected months ago and never used once. Book a free Consultation: 🔗 https://go.genovaflow.com/skool-stop-hitting-claude-codes-token-limit-11-free-fixes-call

Hey everyone! I’m Alex, and I’m excited to join the AI Automation Flow . I’m really interested in how creators are growing newsletters, building engaged audiences, and turning subscribers into real business opportunities. Looking forward to learning from everyone here and connecting with other newsletter builders. What’s one thing that has made the biggest difference in growing your newsletter?