GPT-6 gets much cheaper

GPT-6 is getting cheaper to use

OpenAI cut API prices in half.

GPT-6 Sol now costs $2 per million input tokens and $10 per million output tokens, while Luna drops to $0.10 and $0.50.

The performance numbers matter too. On AutomationBench, Sol scored 33.2% at $0.27 per task, while OpenAI says Claude Opus 5 at maximum effort cost 11.1 times more per task and scored 26.9%.

I think price is the bigger story. As agents sit at a computer working through longer coding and business tasks, the bill becomes part of whether teams actually keep them running.

OpenAI is also improving prompt caching, with cached input reads getting a 90% discount. GitHub says the changes have already cut fresh prompt processing by more than 50% across billions of requests.

That could change which model developers reach for every day.

How much intelligence is enough when the cheaper model already does the job?

Google wants Gemini under your cursor

Google is selling the hardware now.

Its new $899 Googlebook puts Gemini directly into everyday laptop actions, including the cursor, dictation, and small widgets, with models from Acer, ASUS, Dell, HP, and Lenovo.

Magic Cursor lets you hover over an email and ask whether it looks suspicious, while Rambler turns spoken brain dumps into cleaner text. You can picture someone talking into the laptop while an unfinished document sits open on screen.

I’m skeptical these features alone will convince many people to replace a working laptop. Most of Gemini’s useful capabilities can already live inside software, and increasingly capable computer agents make an AI-powered cursor feel less important.

The bigger opportunity is distribution. Google says roughly 50 million Chromebooks are used in schools, and some existing machines could eventually transition to the Googlebook experience.

That gives Google a direct route for making Gemini part of how millions of students learn to use a computer.

Will people buy AI hardware, or simply expect every laptop to gain these features?

AI labs are spending heavily on better data

Snorkel AI just tripled its valuation.

The company raised $350 million at a $3.5 billion valuation, only 17 months after its previous round, while its annualized revenue run rate jumped eighteenfold to $375 million.

The interesting part is what Snorkel sells now. It moved beyond data-labeling software into finished training datasets and reinforcement learning environments, using software-generated data alongside human specialists.

I think that shift explains the valuation better than the broader AI boom. Labs can buy more chips, but improving advanced models also requires increasingly specialized examples, evaluations, and environments built around difficult tasks.

There are people behind those datasets, working through screens and domain-specific material. Rivals like Mercor and Handshake are already reporting gross annualized revenue in the billions, although much of that money goes directly to specialists.

The open question is how long demand keeps growing this quickly as models get better at generating their own training data.

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.