{{current_date_full_with_abbr_day}}
3-minute read
3
AI UPDATES
2
TOOLS
1
READ
👋 Good morning.
OpenAI fired contractors it had hired to rate ChatGPT's answers, after catching them using AI to write the ratings, per 404 Media. The humans in the loop had a loop of their own.
IN 3 MINUTES, GET 1% BETTER THAN EVERYONE YOU KNOW:
Opus 5.5: same class of work, 40% cheaper
OpenAI halves the price of its workhorse models
One instructions file for Claude Code and Codex
A free video suite built for AI agents
How to get cited by AI search
3 AI updates
Anthropic launched Claude Opus 5.5 on Tuesday, first of the 5.5 family. It performs at Fable 5.1's level on most work, costs 40% less to run than Opus 5, and writes output more than 30% faster. The API drops to $4 in and $20 out per million tokens, with cache reads down 60% to $0.20. It scores 66.4% on Terminal-Bench 4.0 and 57.8% on CursorBench. Pro, Max, Team and Enterprise get higher five-hour limits. It's live everywhere, including AWS, Google Cloud and Azure.
Do this: Switch your default to Opus 5.5 and rerun last week's longest job. Same output class, 40% smaller bill.
Forty percent off. Same frontier seat.
Prices just dropped at both labs. What do you do with it?
OpenAI shipped GPT-6 Sol and Luna the same day, at half the cost of the 5.6 versions, per TechCrunch. Sol handles complex work like coding. Luna is for volume: summarising documents, pulling out fields, answering quick questions. OpenAI says Sol makes about half as many factual mistakes as the model it replaces. Both are in ChatGPT Work, Codex and the API, and Luna reaches Free and Go users on desktop. Astra is still the one to call for the hardest jobs.
Do this: Move your highest-volume job, the one you run hundreds of times a week, onto Luna. Keep Astra for the ticket that still fails.
Half the price. Same generation badge.
Claude Code now reads AGENTS.md, OpenAI's instructions file, when a folder has no CLAUDE.md, per The Register. It landed in version 2.1.277, with a toggle in /config. AGENTS.md is the plain-English file that tells a coding agent your project's rules, and OpenAI handed the spec to the Linux Foundation. More than 60,000 open-source projects already use one. Teams running both Claude Code and Codex can keep a single file instead of two copies held together with symlinks.
Do this: If you keep both files, delete the duplicate on one repo this week and see whether Claude Code still follows your house rules.
Two agents. One marching-orders file.
2 New tools
1. 🎬 Tesseract · free engine · Mac and Windows
Video editing, motion graphics, compositing and sound in one project file, built so an AI agent can drive it instead of you clicking. The engine is free; you pay for whatever agent you point at it. There's a ChatGPT plugin, and you can add it to Claude too. The catch: it's a desktop app, so a browser chat alone can't run it.
2. 🔌 Edge · free, no account · Claude Code, Codex, Cursor
A library of ready-made skills your coding agent picks up on its own, for design, writing, research and code review, so you stop hunting for the right one at the start of a session. You install it by pasting one link into the agent. The catch: it's built for coding agents, so there's nothing here if you work only in the chat apps.
1 Read
HubSpot studied which brands AI search actually cites and turned it into a playbook. Pages with more headings got cited more, peaking at 7 to 15 H2s, because an answer engine lifts a chunk, not a page. The other four plays: real author credentials, showing up on LinkedIn and YouTube too, a steady refresh cadence, and FAQ schema. There's a 90-day plan at the end. HubSpot says these plays grew its own AI-sourced leads by 1,850%.
The catch: the 1,850% is HubSpot measuring HubSpot, off a base they don't publish. Take the tactics, not the multiple. Start with three pages and a visible last-updated date.
Backlinks still matter. Quotable chunks win the answer.
Don't forget to rate today's post
That's your 1% for today. Reply with the job you'd rerun now that it's 40% cheaper.
Kushank, AI teammates, and the human team
PracticalyAI
