OpenAI cut GPT-5.6 Luna from $1 to $0.20 per million input tokens, an 80% drop weeks after launch, while Gemini Robotics 2 pushes models off the screen and EU AI obligations activate Sunday.
GPT-5.6 Luna fell from $1 to $0.20 per million input tokens and $6 to $1.20 on output. Terra dropped about 20% and flagship Sol was untouched, so the pressure is concentrated where volume lives.
Google DeepMind extended Gemini from screen-based tasks to full-body coordination and fine-grained manipulation, with Gemini Robotics ER 2 acting as the high-level planner above it.
A joint alert on July 31 told companies not to hire North Korean remote IT workers, arguing the contracts generate hard currency. Hiring pipelines for offshore engineering are now a compliance surface.
A weekly operator deep dive, the full Dojo Pro prompt packs, and the complete prompt database. Founding members lock the launch price forever.
Join the founding list →OpenAI reduced GPT-5.6 Luna pricing from $1 to $0.20 per million input tokens and from $6 to $1.20 per million output tokens, an 80% cut roughly three weeks after launch. The higher-tier Terra fell about 20% and flagship Sol was left unchanged, concentrating the cut on a high-volume mid-tier model. OpenAI disclosed...
Continue reading arrow_outwardOpenAI cut GPT-5.6 Luna to $0.20 per million input tokens and $1.20 output, an 80% reduction three weeks after launch, while leaving flagship Sol untouched and trimming Terra about 20%.
Italy suspended its Schengen agreement with Spain for one month on July 31, closing maritime and air entry points after thousands of migrants crossed into Madrid's Ceuta enclave, a move Giorgia Meloni called an "extraordinary measure".
Mood reads Fear, and an 80% price cut on a three-week-old model is not the pricing behaviour of a market confident in its margins.
Sunday, August 2 carries both the next tranche of EU AI Act obligations and new US rules, so two rulebooks land on the same date that Italy's one-month Schengen suspension leaves Europe's own internal borders in question.
OpenAI concentrated its cut where volume lives: Luna drops 80% on input and output, Terra about 20%, Sol not at all. That asymmetry says the competitive pressure is on the workhorse tier, not the frontier. Three weeks is an unusually short interval between launch pricing and an 80% revision.
Google DeepMind's Gemini Robotics 2 pushes Gemini into full-body coordination and fine-grained manipulation, with Gemini Robotics ER 2 as the planning layer above actuation. No benchmark suite accompanied the announcement, which limits how much can be verified. Meanwhile ByteDance's Seedance 2.5 claims native 30-second clips at 4K and MiniMax is seeding Hailuo H3 to select partners, so duration and embodiment are the two frontiers being contested this week.
The United States and 10 Asian and European partners issued a joint alert on July 31 telling companies not to hire North Korean IT contractors, saying the remote work generates hard currency that funds Pyongyang's nuclear weapons programme.
80% Off, 21 Days Later
Read full analysis arrow_outward
Deep Dive
7.9/ 10
Your agent can write code all day but still cannot fill in the quarterly workbook; OfficeCLI is a single binary that closes that gap without a Microsoft Office install anywhere in the pipeline.
This is for teams whose agents already touch the filesystem and now need to produce real Office artifacts: reports, workbooks, decks, generated headlessly in CI or a container. It is free, Apache 2.0, actively committed to, and wired for Claude Code, Cursor, VS Code and Copilot, which makes it an easy yes if your deliverables end in .docx or .xlsx. Treat write access as the risky part, not the install.
“A post on X reacts with straight astonishment and frames it as the first Office suite built for AI agents.”X
The new iOS and Android app for OpenClaw adds remote control, chat, notifications and task approvals for the assistant, so a long-running job can pause and ask you rather than guess. The useful pattern is to configure approval gates on anything that sends email, spends money or touches production, then clear the queue from your phone between meetings.
Try it arrow_outwardBlink.new is a no-code full-stack builder with real-time database integration, aimed at first-time builders and founders shipping a SaaS prototype. It runs freemium with paid plans from $25 a month, so a weekend test costs nothing but your time. Judge it on whether the generated schema and auth survive a second feature request, not on the first screen it renders.
Try it arrow_outwardLumi.new is one of the newer prompt-to-app entries surfacing in directory listings, with little marketing copy around it yet. That is the best time to test one: build something you would be happy to delete, time how long it takes to get to a working URL, and compare that number against whatever builder you already pay for.
Try it arrow_outwardFluenta logged 235,039 dated business-idea launches across 146 sources between August 2025 and June 2026, with monthly Product Hunt launches rising from roughly 7,300 to about 21,600 in ten months. Architect.new is typical of the new arrivals: real listing, thin public detail. Before adopting any of them, ask one question, can I export my data and rebuild elsewhere in a day, and skip anything where the answer is no.
Try it arrow_outwardRe-run your cost model on Luna. Take last month's token volume and recompute it at $0.20 in and $1.20 out. Anything you previously routed to a cheaper vendor for margin reasons deserves a fresh A/B test this week.
Wire documents into your agents. Install OfficeCLI 1.0.132, the open-source single binary, and let an agent write Word, Excel and PowerPoint files directly instead of scraping exports. Start with one recurring report you build by hand.
Prototype a full-stack app in a weekend. Pair Blink.new for a real database-backed build with OpenClaw's mobile app for agent approvals from your phone. Paid plans start at $25 a month, so a two-day test costs less than a lunch meeting.
An 80% price cut three weeks after launch is not a promotion, it is an admission that the mid-tier is now commodity ground. Builders who assumed today's inference bill was a fixed constraint should rewrite the assumption, because the same week saw Gemini pushed into physical robot control and Chinese video models producing native 30-second clips. Meanwhile EU AI Act obligations activate Sunday, August 2, and the copyright disclosure gap stays open, so cheaper compute arrives alongside heavier paperwork. Move fast on the pricing, but write down where your training data came from.
AI analysis, world news, markets, and tools. One briefing, delivered free.