OpenAI is REALLY far ahead right now.
Arena just updated its Alignment Index, and OpenAI holds ALL seven top spots. Fable 5.2 debuts at #8. Opus 5.5 went from #6 to #17 under the new method. The best open model is Kimi K3.1 Pro at #9.
It measures how agents behave in real sessions: doing things you didn't approve, saying work is done when it isn't, and crediting you with things you never said.
Claude may still write the best code. On behaving, nobody is close to OpenAI.
This is so true
I'm convinced that Fable lovers under specify their prompts and Anthropic figured out that if they bake a little randomness into their output, people will mistake that as genius creativity.
Early testers tend to sling a lot of huge token burning tasks at models because they get a short window of infinite usage. This is not the typical experience of actually building something. Astra has proven AMAZING to me, but I tend to yap long instructions to build what I want with intention. And so when you just ask for "build me a new expansion in Majora's Mask" in Astra, you'll get a lukewarm return to the mean "Bellwhether" level.
To me, this is a symptom of OpenAI building tools vs Anthropic building a being. And it's sad to me that there was a strong day 1 preference for the model that rewards pure "pull the lever and see what comes out" style prompts (Fable.) That's technically cool, but it also represents total human replacement to me.
I prefer the tool to power me up instead of letting it take over. The voices I've seen discuss Astra as being KILLER for research, math, science, which are inherently more goal oriented. The kind of tasks that the user is driving rather than taking a ride.
Being real though I think we reached crossover months ago where you can use any frontier model to accomplish any task, including wild stuff outside of your domain of expertise. So use fable. Astra. Neon!!! They are all immaculate and complaining about anything today at the peak of history is pure wasted breath
1
1
78
For the first time since we launched Design Creativity Bench, we have a new SOTA!
GPT-6.2 Astra is #1 with a creativity score of 70.4, followed by GPT-6.2 Sol at 70.0.
The bigger story though, is how this generation's Sol has caught up to Astra in terms of creativity.
GPT-6.1 Sol was 1.8 points behind GPT-6 Astra. With 6.2, that gap has narrowed to just 0.4 points, while Sol is 4.5x cheaper per design. Both models also now sit on our creativity-vs-cost Pareto frontier.
The lead over Anthropic's models has also grown. GPT-6.2 Sol scores nearly 20 points higher than Claude Opus 5.5 and Claude Fable 5.2, compared with GPT-6.1 Sol's 13-point lead. It also costs 4x less than Opus 5.5 and 6x less than Fable 5.2.
For UI generation, this is a big step forward. Astra sets a new bar for design creativity, while Sol gets remarkably close at a fraction of the cost.
Congrats to @sama, @thsottiaux, and the @OpenAI team!
Explore the full leaderboard: kombai.com/benchmark
Ddox retweeted
Everyone is discussing whether GPT-6.2 beats Claude.
Wrong question IMO.
On Artificial Analysis, Claude scores 65 and costs $26.54 a task. OpenAI's new model scores 62 for $4.
$4 DOLLARS!!!!!!!!!!
That means OpenAI is going straight for enterprise budgets. CFOs don't buy the smartest model, they are business people and the one that's good enough at a sixth of the price will win.
That's Anthropic's whole business during it's IPO.
I'd be scared if I were them.
Ddox retweeted
Replying to @jessexai_
One could have read the same teleprompter script 120 years ago, and said the same thing about farmers and tractors. One would have been equally wrong. We have a new tool, it obsoletes old labor-intensive ways to do things, and people adapt to do more productive things.
Ddox retweeted
Really proud that the new Astra for Law (and 6.2 Astra itself) leads the industry in eliminating hallucinations:
20x fewer hallucinations than Opus 5.5, and 25x fewer than Fable 5.2!
Especially important for law firms, which are particularly sensitive about hallucinations.
Ddox retweeted
Tested GPT-6.2 Astra & Sol. OpenAI Hate is FORCED...
Theo and Matt Shumer raised valid criticisms, but I disagree with how harshly they're judging this release.
The models have real shortcomings, but also some impressive improvements that deserve recognition. My Video will cover this in 1-2 hours.
hate against openai is forced
it’s been about 2 hours since gpt 6.2 released and all i see on x is hate
> this is an opus 5/4.8 level model
> there is nothing good about these model
> why even release a models this bad
> and a lot more too
got access to both models in codex 1.5 hours after release and from my testing so far, they’re huge step up from 6.1 sol and 6 astra
models are super cheap and efficient too as 6 astra used to burn through my plus limits in just 30 mins
but with 6.2 astra, i’m using on xhigh and still have over 80% limits left
to me all this hate on x feels forced. maybe even coordinated attack
Ddox retweeted
Every influencer today: "I HATE Astra 6.2"
Every normal person today: "Astra 6.2 is blowing my mind"
think for yourself
this
Hi friends, in the latest weeks I worked with GPT 6.2 Astra. With it I built Redis features and most of YouRec (testflight.apple.com/join/N4…). My feeling is that it is the *best* model for system programming in C and for complex data structures. (continue)
1
63
Ddox retweeted
Over the past week, I’ve been overwhelmed by the love and support of friends, industry colleagues, and people I've never met. Thank you to everyone who has shown kindness to my family and me. It means the world to me.
I came to the United States 53 years ago, and becoming an American citizen is one of the greatest gifts and proudest moments of my life. NVIDIA has been my life’s work. Building it into the company it is today alongside my colleagues has been a privilege and an honor. Nowhere but in America could a company like NVIDIA be possible. I love our country, and I love NVIDIA.
I did not live up to the standards expected of everyone at NVIDIA, and I have made serious mistakes in my personal life. I created problems for the Board that I deeply regret. I own my mistakes completely, and I am committed to cooperating with the Board’s independent investigation.
Since my departure, I’ve seen an outpouring of calls from NVIDIA’s employees, investors, business partners, and customers for me to return as CEO. Many continue to reach out to me directly as the company navigates complex technology, market, competitive, and geopolitical dynamics. NVIDIA is a once-in-a-generation company. If the Board of Directors decides that my return is in the best interest of our shareholders, our employees, the company, and America, I would be honored to answer these calls and return as CEO.
Thank you again to everyone who has shared their good wishes. I am profoundly humbled by your support. To my family: I love you, and I will spend the rest of my life being worthy of yours.
Jensen
Ddox retweeted
Here is my most ambitious demo to date, built with GPT-6.2 Astra. A 1:1 repro of the map of Paris. Walkable, drivable, with traffic, bus, pedestrians, bikes and more than 50 3D-modelled landmarks. It's a mix of authored assets and procedural landmass and texture generation.
Ddox retweeted
Lmao bro gpt 6.2 is actually good. Both Astra and sol are doing great, reminds me of 5.6 honestly. Yes it nuked my usage but it’s fine. Not sure what is up with some of the big peeps like matthew, theo and some other folks talking shit about it. gpt 6.2 is not bad at all
Bro don’t tell me this is some strategy to downplay openai just because anthropic ipo is on Thursday, but it’s getting pretty obvious now isnt it
GPT-6.2 Astra from @OpenAI on ARC-AGI (Verified):
- ARC-AGI-3: 100.0% (SOTA), $4.1K (provider adapter), 70.3%, $24.2K (standard)
- ARC-AGI-2: 95.8% (SOTA), $0.50/task
- ARC-AGI-1: 98.5%, $0.17/task
GPT-6.2 Astra is the first model to achieve a 100% score on ARC-AGI-3.
Ddox retweeted
A sad day for NVIDIA, developers, America, and all humanity. Jensen founded and ran one of the few companies to reach a massive scale while driving massive continued innovation and ubiquity, while others fell into the trap of ever-increasing extraction.
Ddox retweeted
Jensen Huang rescued me from probably living in debt the rest of my life. I was told I was one of just three hires that he approved during the pandemic, during a hiring freeze. I've never spoken to him, and only passed by him in HQ once. His personal life is frankly none of my business. But as a leader, I have yet to meet a better one. Ousting him is financial suicide, and I can only assume there is a hidden ulterior motive because I can't fathom any other reason for destroying the company's value.
Ddox retweeted
Reality of our days:
1. I am sitting on a set of vulns (4+) that allows me to remotely take over a consumer home automation device I have (3+ diff ways).
2. The vendor doesn't only serve consumers, but also critical infra (announced partnerships include pharma manufacturing plants, airport logistics, hospitals, etc) across various locations in Europe and the US (not the same product line, ofc, so maybe moot).
3. Claude and Astra refuse to help me explore even my own device, so now I'm using Chinese hosted models for investigating.
4. The vendor (with a revenue of $0.75B) does not maintain any security disclosure practice whatsoever, nor does respond to emails.
So, to recap, I need to outsource vulnerability hunting of (potentially, I don't know) critical western infrastructure to Chinese models, because the western models refuse; and the vendor doesn't give a shit. Fun timeline.
PS: The vendor is Swiss.
























