Sam Altman once described a future where we buy intelligence through a meter, much like electricity.

We already do.

Your subscription buys a limited amount of intelligence every month. APIs charge by token. When the limit ends, the intelligence stops.

A large company can spend far more agent runs on the same problem. A student or solo builder has to work inside a much smaller limit.

That creates an intelligence gap. It does not tell us whether the gap will last.

So far, powerful models arrive expensive and become cheaper. Open models copy much of the capability. The people who were locked out eventually get access.

Recursive self-improvement could break that pattern.

I am Alex, welcome to ShortCu8 by Innov8.

Lets Dive Deep 🐰

Watch the catch-up time.

If open models reach the old frontier while the new frontier advances at a normal pace, the gap remains temporary.

If one model starts improving its successors faster than everyone else can follow, the gap can grow after every generation.

A frontier model usually enters the market behind a high price, a subscription limit or restricted access.

Then researchers find cheaper training methods. Competitors reproduce parts of the capability. Smaller models improve. Open-weight releases let more companies work from the same starting point.

We can see that compression happening now.

On July 30, OpenAI cut GPT-5.6 Luna's API price by 80%, bringing it to $0.20 per million input tokens and $1.20 per million output tokens.

DeepSeek's updated V4 Flash API costs $0.14 per million cache-miss input tokens and $0.28 per million output tokens.

Moonshot went further and released the full weights of Kimi K3.

These models cannot be compared cleanly from price alone. They still show the direction: useful intelligence is losing its launch price quickly.

A closed model keeps its capability behind one company's price, permission and product rules.

Open weights let an entire ecosystem study the model, modify it and build competing services around it. One release becomes a starting point for the next round of research.

Open models do not erase the frontier lead on launch day. They shorten the period when only the frontier company has that level of intelligence.

This is why today's intelligence gap still looks temporary.

Recursive self-improvement, or RSI, happens when AI helps create a stronger AI, which then becomes better at improving its own successor.

OpenAI already measures progress toward this. Its GPT-5.6 release includes an internal RSI Index covering research debugging, kernel optimization, training experiments and improving another model. OpenAI says GPT-5.6 Sol scored 16.2 points above GPT-5.5 on the combined evaluation.

OpenAI also used GPT-Red to find attacks that were later incorporated into GPT-5.6's safety training.

Then an internal model called Astra produced a 249-page collection of ten results in mathematics and theoretical computer science.

These are signs of AI-assisted improvement. Full RSI would go further:

There is no public proof that any company has closed this full loop.

Assume OpenAI closes the loop one year before everyone else.

Other labs begin trying to match Model A. During that work, Model A helps OpenAI create Model B. Model B speeds up Model C.

By the time an open model catches Model A, OpenAI may already be using Model C or D to produce its next generation.

The open ecosystem would still receive powerful intelligence. It would keep chasing an older frontier while the leader's research speed compounds.

Distillation cannot reproduce private experiments, training systems, failed approaches and unreleased models. A one-year advantage could therefore grow instead of expire.

For that to happen, the RSI loop must be reliable, largely autonomous, exclusive to one group and faster than every competitor's catch-up process. We have evidence of AI assisting research. We do not have evidence that all four conditions have been met.

We do not have a permanent intelligence underclass today.

Open models keep reducing the gap. Falling inference prices keep lowering the cost of useful intelligence.

But if one company has already closed the recursive self-improvement loop, the temporary gap could become permanent.

Watch how long it takes open models to catch the frontier.

If the delay keeps shrinking, open models are winning.

If it grows across several releases, that may be our first public sign of full RSI.

Now go build something great!

We read every reply - just reply to this email and let us know how we can improve !

#AD1

The best assistants don't multitask their attention across a hundred tools. Neither does Catch. It's an AI agent that focuses on one thing — the admin work you'd rather not touch — and does it exceptionally well.

Scheduling, flights, restaurants, follow-ups, vendors, clients. You hand it over; Catch handles the back-and-forth and comes back with it done.

No context-switching. No dropped balls. Just your admin, quietly cleared — so your focus stays on the work only you can do.

Meet the agent built for admin, and it'll be ready to work before your next meeting.

Get started at catchagent.ai — and give your attention back to what matters.

Start now for free

#AD2

Most e-commerce sellers juggle 6–8 tools and pay hundreds monthly to keep operations running. StoreClaw replaces the stack with one autonomous AI engine that monitors competitors, optimizes listings, automates marketing, and tracks profit 24/7. Connect your store and let AI handle the work — no prompts, no complex setup, no credit card required.

Try if for free today