GPT-5.6 debuts this month, with agentic coding reportedly surpassing Anthropic’s Mythos! Three flagship models converge in the same June, as two AI giants simultaneously race toward IPO—yet Altman has introduced an even bigger internal variable: if AI learns to improve itself first, going public isn’t urgent.
Article author and source: AI World
GPT-5.6, coming this month!
Just now, OpenAI launched an unexpected series of moves.
The familiar model identifiers used by ChatGPT have been completely replaced with Intelligence "Intelligence Levels."
The Wall Street Journal exclusively reported that OpenAI is preparing to significantly lower its API pricing in preparation for a price war with Anthropic.
Immediately following, Chief Scientist Jakub Pachocki personally announced that the new model, codenamed 5.6, which significantly surpasses its predecessor, will be launched this month.
Price cuts, redesigns, new models—a crazy Wednesday.
But all of this combined doesn't compare to a single offhand comment Otterman made inside Slack—
If the recursive self-improvement of AI accelerates quickly, the benefits of delaying the launch increase further.
An $852 billion IPO; Altman says there's no rush.
The current broader context is that everyone is rushing to go public.
Anthropic secretly filed an S-1 with the SEC on June 1; SpaceXAI is already on its roadshow with a $1.77 trillion valuation. OpenAI followed suit with its own filing on June 8. Combined, the three companies are valued at approximately $3.6 trillion—equivalent to France’s entire annual GDP.
Investment banks have consistently advised that the first company to go public will define the valuation framework for investors in the AI sector.
First-mover advantage is a strategic imperative.
Yet at that moment, Ultraman introduced a variable that no one had publicly discussed:
The faster the AI's recursive self-improvement takes off, the greater the benefit of delaying the IPO.
Because technology and the world may change in unexpected ways, there could be good reasons to be a private company during that time.
He didn’t mean “doesn’t want to go public”—rather, once AI reaches a critical point of self-improvement, the entire rules of the business world could be overturned. At that point, private companies would have far greater flexibility than publicly traded ones.
Anthropic's data indirectly supports this judgment.
Their internal reports show that the time it takes for AI to complete tasks is doubling every four months, and engineers' quarterly code output has surged to eight times the previous level.
Seven weeks per cycle—how far is RSI?
On the same day Ultraman made this statement, his chief scientist was demonstrating through action that day might be much closer than imagined.
GPT-5.4 was released on March 5, and GPT-5.5 followed on April 23, six weeks later.
GPT-5.6 is scheduled for June, again following a 6 to 7-week cycle.
This is a steadily accelerating curve, with no sign of slowing in the leaps of capability between generations.
Overseas communities have long been thoroughly investigating the alleged leak of GPT-5.6.
Since mid-May, developers have detected routing traces of GPT-5.6 in the Codex backend logs, internally codenamed iris-alpha.
Subsequently, ember-alpha and beacon-alpha appeared, followed by kepler and kindle.
By early June, kindle-alpha was confirmed as the current release candidate.
Someone discovered the anonymous model "Kindle" on Design Arena, and after running several rounds of tests, concluded that it is the public test version of Kindle-alpha.
Later, Kindle was removed, but the existence of GPT-5.6 was already set in stone.
Currently, the community is most focused on improvements in two areas.
The first is frontend generation capability. The model can directly output clean, production-ready UI interfaces without requiring complex prompts.
A leaker used the earliest iris-alpha checkpoint to generate a note-taking app called Lumen Notes, with no guidance—featuring a lavender color scheme, grid-aligned layout, clear hierarchy, and appearing just like a screenshot of a mature SaaS product.
The second is agentic coding capability.
Renowned developer Mark Kretschmann stated on 𝕏, "As far as I know, GPT-5.6 is extremely powerful and outperforms Anthropic's Mythos on multiple agentic coding benchmarks."
In recent events, Ultraman stated that enterprise clients are becoming increasingly sensitive to the cost of using AI.
Therefore, price may be one of OpenAI's most critical variables going forward.
Anthropic has just released Fable 5 and Mythos 5, with API pricing at $10 per million input tokens and $50 per million output tokens, approximately double the current Opus pricing.
Meanwhile, GPT-5.5 is currently priced at $5 and $30, already half the cost.
Moreover, according to WSJ, OpenAI is even considering further significant price cuts to proactively engage in a price war with Anthropic.
If GPT-5.6 brings both enhanced capabilities and a price reduction, it’s a one-two punch for Anthropic.
From Choosing a Model to Choosing Intelligence
Meanwhile, the product team has also been busy.
On June 10, Adam Fry, Head of Products at OpenAI, announced on𝕏 that the ChatGPT model selector has been officially revamped and is rolling out globally to Plus and Pro users.
Previously, when you opened ChatGPT, you were greeted by a long list of model names.
Thinking-Light, Thinking-Standard, Thinking-Extended, Thinking-Heavy, plus Pro Standard and Pro Extended—six or seven options crammed together, instantly overwhelming with choice anxiety.
Now all of that has disappeared, leaving only one word: Intelligence.
Six tiers are listed in ascending order: Instant, Medium, High, Extra High, Pro Standard, and Pro Extended.
In other words, it has shifted from “which model do you want to use?” to “how smart do you want the AI to be?”
Thinking-Light has been removed, as less than 1% of paying users utilize this tier. Thinking-Standard is now called Medium, Thinking-Extended is now called High, and Thinking-Heavy is now called Extra High. Pro Standard and Pro Extended retain their names but have been moved into the Pro submenu.
Replace the model every 7 weeks. Update the product interface on the same day. Prepare for a price reduction on the same day.
Every acceleration signal is making Ultraman’s statement about the RSI seem less like a hypothesis and more like a prediction.
The faster, the more this sentence sounds like a prophecy.
Once AI learns to improve itself, the priority of going public may need to be reassessed.
Within 24 hours of him making that statement, Anthropic’s Claude Fable 5 topped the new Agent Arena leaderboard, achieving the largest margin in the leaderboard’s history with a net overall improvement of 11.2%, leaving GPT-5.5 in fourth place.
In June, three flagship models collided head-on: Fable 5, Gemini 3.5 Pro, and GPT-5.6, all competing in the same set of capabilities—reasoning, coding, agents, and frontend generation.
But the real competition may not be at this level.
Whoever goes public first gets Wall Street’s capital. Whoever achieves RSI first gains the power to rewrite the rules.
The advantage of the former is measured in years, while the advantage of the latter may be measured in days.
Once a company’s AI truly completes the self-improvement loop, its lead will expand exponentially, and no amount of additional funding will allow latecomers to catch up.
This is probably what Ultraman really meant: IPO is the means, RSI is the end goal.
GPT-5.6 is for competitors to see, the price cut is for enterprise customers, and the RSI section is for history.