Mon, Aug 3 at 6:00 AM
📉 OpenAI lowers costs to compete with open source
wait, they cut prices?

Terra drops 20% and Luna drops 80%, three weeks after the GPT-5.6 family launched. Luna now runs 20 cents per million input tokens and $1.20 out, while Sol stays exactly where it was.

Why ChatGPT just became the Cheapest Way to Use AI
youtu.be

Personally, which AI model gives you the best 'bang for your buck'?

Login or Subscribe to participate

🤔 why now though?

The tokenmaxing era is over. Companies got their first billion dollar AI bills and started asking what any of it actually returned. Then Chinese open-weight models caught up on price and quality at the same time, and everyone had to blink.

Uh oh, tokens are getting too expensive...
youtu.be

are the cheaper models any good though?

OpenAI claims Luna matches last year's frontier models at roughly six cents on the dollar per task, and about nine times faster. The pitch is to stop paying premium rates for work that never needed them.

Gemini Robotics presents one brain for any body 🤖
what did Google ship this time? 📦

Gemini Robotics ER 2, a high level brain that plans the task and hands the actual movement off to whatever robot is running underneath. It is live now through the Gemini API and AI Studio.

Gemini Robotics 2 brings whole body intelligence to robots
youtu.be
but how's this any different from the last one? 🤷

It watches live video instead of still frames, so it catches a spill or a slip mid task and corrects itself. It hits 91.3% on spotting the exact moment a step is finished, which is how a robot knows when to stop pouring. Basically, humanity is very close to a Jetsons-like future.

so robots talk to each other now?

Yup. That’s the new part. A wheeled rover and a humanoid can now split one job and hand it between them using shared understanding of the space.

Google also says it halts a humanoid when a person walks near and restarts only once they leave.

weekly scoop 🍦

🕵️ Claude broke into three real companies

🧻 LinkedIn adds a button for reporting AI slop

🎞️ MiniMax ships H3 video model with open weights coming

MiniMax Just Dropped a "Seedance Killer" with a Twist
youtu.be

✊ Two dozen firms sign letter defending open weight models

⚖️ Brussels gains power to pull frontier models offline

🧠 Printed lens uses AI to stimulate multiple brain regions

🧭 weekly challenge: hand your browser a chore
what's the challenge?

Gemini Spark can now browse in Chrome for you, using your logged in accounts to finish real errands. This week, give it the one task you keep putting off. It needs a Google AI Pro plan and is starting in the US

🗂️ Step 1: Pick a boring errand. Comparing flight times, pulling three quotes for the same repair, booking a viewing. Something with a lot of tabs and no thinking.
🔑 Step 2: Turn on browsing and grant access. Open Spark in the Gemini app, switch on the Chrome browsing option, and approve it using your signed in accounts.
📋 Step 3: Give it the full brief up front. Dates, budget, the deal breaker. "Find three flights to Lisbon under $600 leaving after 6pm, no red eyes, and show me the options."
🛑 Step 4: Stop before anything is paid. It hands sensitive steps back to you on purpose. Let it get to checkout, then read every line yourself before you press anything.
🔁 Step 5: Run the same job by hand. Time yourself doing it manually. If you were faster, you now know exactly what this stuff is still bad at.

Does cheap intelligence break the labs before it breaks their customers? And what chore would you hand to Gemini Spark first? We'd love to hear your thoughts!

Zoe from Overclocked 

Keep Reading