GPT 5.5 Codex 90M Downloads Beats Claude 12x Team Saves 32K Monthly
Editor’s Note: Sam Altman himself called GPT-5.5 a “loner genius.” A 16 person team canceled Claude overnight and switched to GPT-5.5, saving 32,000 dollars per month. Codex hit 90 million downloads in one week, crushing the rival by 12 times. The great developer migration has begun.
Just now, Sam Altman gave GPT-5.5 a nickname that set the whole internet on fire. He called it the “loner genius.”
He could not help but say, I can not believe we actually built an AI like this.
GPT-5.5 has been live for half a month. Altman has shared his excitement many times on social media without holding back.
In Altman’s own words, GPT-5.5’s “raw smarts” has opened a gap that no one else can cross.
Test scores crush the rest. Token use is tiny. Power and beauty are at full blast.
Top AI minds are voting with their feet for GPT-5.5. AI professor Derya Unutmaz even said it could straight up be called GPT-6.
Also today, a chart is going viral across the web. Codex powered by GPT-5.5 saw its downloads shoot up in May to 86.1 million, far above Claude Code.
In just one week, downloads broke 90 million. That is 12 times more than Claude Code.
At the same time, feedback from developers is wildly proving the same point.
Many people openly said that GPT-5.5 in real coding tasks has already passed Claude Opus 4.7, most of all in token use.
For the same task, GPT-5.5 uses nearly 40 percent fewer tokens than Claude.
You have to admit, the “loner genius” label hits the mark so well it almost hurts.
16 Person Team Drops Claude Saves 32,000 Dollars Per Month
Morgan Linton, founder of the startup Bold Metrics, posted a message. His tone was calm but the content was explosive.
Officially saying goodbye to Anthropic.
For my 16 person small engineering team, the Codex plus Cursor combo has fully replaced our old setup.
The reason is simple and blunt. Claude Code costs too much.
By contrast, with GPT-5.5 behind it, Codex has lately shown amazing results. And its token use is super high, which means big cost savings.
In real daily work, Bold Metrics still uses Cursor a lot for code review.
The key point is that the team has never hit a usage cap with Cursor. Its built in Composer 2 feature is enough for almost all dev needs.
On Claude token costs, Linton did the math.
Each engineer burns over 2,000 dollars per month. For 16 people that is more than 32,000 dollars per month.
After switching to Codex plus Cursor, GPT-5.5’s token efficiency made costs drop off a cliff. And the power did not drop at all.
What stings even more is his guess. More and more engineering leads will post similar choices.
You have to say, this post hit like a depth charge. It went straight to Anthropic’s weak spot. The product is good, but token costs feel like robbery.
On the Codex side, the numbers already tell the story.
90 Million Downloads in One Week Codex Becomes a Legend
TickerTrends data shows that as of May 3, Codex downloads reached a jaw dropping 86.1 million. Week over week growth exploded by 1,397 percent.
By May 8, that number climbed further to 90 million in a single week.
At the same time, Anthropic’s Claude Code had 7.2 million downloads in the same period. Week over week it actually fell by 38 percent.
One is flying high. The other is bleeding out. The speed of this shift is breathtaking.
The trigger for this growth wave is clear.
On April 30, Codex dropped version v0.128.0. It brought in the lasting /goal workflow that supports multi step task planning across chat sessions.
Plus GPT-5.5’s million token context window and 40 percent token efficiency boost. Developers are voting with their feet faster than any review could show.
Altman himself said in an internal note that he had one word for Codex growth. Crazy.
Microsoft VP Omar Shahine could not hold back his praise. He said Codex is amazing at building Swift iOS apps.
He used just one prompt. Codex built the app straight away and solved 95 percent of the work. Much better than Claude Code.
Right after that, OpenAI developer experience lead Romain Huet said.
Codex can design the UI, write Swift code with GPT-5.5, and run the app in a simulator without even opening Xcode. It can even use “computer control” to click around and do tests.
Developer Dimitris Papailiopoulos also said he clearly trusts Codex more now.
Other developers are also showing their trust with real time praise.
ai video generator porn
ai nudifierLate Night Coding Screen Glows and Data Vanish
Also today, a new topic started trending on social media. Everyone is asking, what exactly did OpenAI change in this model update?
For a moment, the model architecture debate was on fire.
One Word Explains It All OpenAI Built a Real Wall
Former OpenAI researcher Will Depue’s reply said it all.
The gap between GPT-5.5 and Claude is not about being smart. It is about “taste.” That is the real wall.
He gave one example. When a math teacher asks a question, GPT-5.5 will write out the full proof and formula like a top student. Claude, on the other hand, acts like a math teacher’s trusted aide. It may skip steps you already know, which feels more natural but can also feel less solid.
Some people joked that OpenAI’s data team is too robotic. They should learn from Anthropic about the model’s “taste” and “personality.” The gap is about 30 percent.
But the stronger model still wins on “loner genius” power. That is the hard truth.
After all, ChatGPT is a tool. It follows your orders and writes what you ask. That is its job.
On this point, a former Anthropic staff member also agreed. He said he could see the new change right away.
Fresh Brand Colors
Just today, the OpenAI official account posted a set of super techy color images. Big changes are coming.
The old “Ask ChatGPT” button is now “Message ChatGPT.”
And one line says “Call me maybe.” It seems OpenAI is trying to push the product from a tool to a real companion.
Next week, more news is coming.
