sarvesh kapre

Engineer @OpenAI. Prev - @LinkedIn @Atlassian

[sarvesh kapre @_cyberhusky]

GPT-5.6 Sol + ChatGPT Sites turned a weekend idea into something real and then kept pushing it beyond what i had imagined - ran for 320+ hours using /goals.
https://t.co/8YLKAHksoE

this started as a weekend vibecoding project as i wanted one page to explore the infrastructure that powers, computes and connects the world.

with very little design direction, Sol kept improving the interface around the data - power plants, grids, data centers, submarine cables, internet exchanges and now satellites!

it has grown into something closer to a living model of global infrastructure than the simple map i originally imagined

[Sachin Tendulkar @sachin_rt]

I’m excited to announce my partnership with OpenAI. We have some interesting things coming up, and I can’t wait to share them with you.

I’ve always been curious... One question usually leads to another, and ChatGPT has certainly encouraged that habit. But it has also proved useful when I’ve needed to get things done, like planning my last trip with the family.

That’s what I’m looking forward to with this partnership. Discovering what else I can do with ChatGPT, sharing what I find useful, and encouraging more people to give it a try with something they care about.

A question, an idea, something you’ve always wanted to explore. There are so many places to start.

Looking forward to exploring together.

#Partnership

[Andon Labs @andonlabs]

OpenAI had a bug, so we reran Blueprint-Bench for GPT-6 Sol and Luna.

Sol improved slightly; Luna made a massive leap!

[sarvesh kapre @_cyberhusky]

Big improvements in Luna since the bug fix.

[SkalskiP @skalskip92]

someone at OpenAI pays attention to my tweets

they just fixed recent GPT-6 release

GPT-6 Luna got faster, cheaper and more accurate

thanks OAI for the weekend fix

↓ examples, leaderboards

[sarvesh kapre @_cyberhusky]

@OpenAIDevs thank you and shoutout @skalskip92
and offcourse @IshaanSingal and team 🙌🏼

[sarvesh kapre @_cyberhusky]

@skalskip92 another banger post
team is taking a look - thank you

[Sam Altman @sama]

We want the OpenAI API to feature the best model at every price point and to be the best at every modality (text, code, image, video, etc).

And then we want you all to come up with great ideas and build them and to get to be happy users. The best ideas will come from you all.

[OpenAI @OpenAI]

Please welcome GPT-6 Sol and GPT-6 Luna to the GPT-6 universe.

GPT-6 Sol and Luna build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to support work at scale.

We’ve also made caching and inference more efficient, and we’re passing the savings directly to you: 50% lower API prices for Sol and Luna compared with GPT‑5.6 promotional pricing.

[Choblin @choblin29]

🚨wtf. OpenAI says its new internal model it began training just 24 days ago has already resolved more than 100 long-standing open problems across most areas of mathematics.

The pace has surprised even OpenAI's own mathematicians.

An independent group including Edward Witten, Timothy Gowers and Martin Hairer is now helping assess the results and coordinate how they are released.

[sarvesh kapre @_cyberhusky]

@skalskip92 Offcourse the work and craftsmanship - both are quite evident.

[sarvesh kapre @_cyberhusky]

Really enjoyed this deep dive - its very well written and thorough - lots of concrete evals, examples, cost breakdowns, failure cases.

fascinating to see general-purpose models increasingly handle problems that not long ago would have called for specialized computer vision systems. And GPT-6 Astra makes that progress feel very real.

Give this a read

[SkalskiP @skalskip92]

GPT-6 Astra is the best vision model we have tested

I put together our results on detection, segmentation, box prompting, counting, reasoning and video

full post: https://t.co/70tEoHqu0k

↓ examples and trade-offs

[CortexAI @cortexairobot]

We tested GPT-6 Astra, Gemini 3.8 Flash, and Fable 5.1 on two real-world robot tasks from our MolmoAct2 real-world evaluations:
1. Fitting a 5 mm pipette tip
2. Loading a microwave

Astra scored highest on both. 🥇

Here are some few surprising things that Astra did to come out on top. 🧵

[sarvesh kapre @_cyberhusky]

@savarlamov congrats and welcome!

[Alex Bodner @AlexBodner_]

I compared SAM3 vs Astra and I'm honestly impressed.

On nouns it is close: "referee", 16/20 (SAM) vs 20/20 (Astra)
On more complex scenarios it is not even close.
- "the players involved in this attack"
SAM returns all the players. Astra returns 4 of 16 players in the frame, each with an accurate role: on the ball, passing option, closing down.
- "the player with the ball at his feet"
SAM 3 hands you all 18 players. Astra returns the one, 42/45.
- "number 10" and SAM3 hands you nothing, Astra is able to read it.

Astra is so good at understanding the context of the image, and this is just the beggining, planning to build very cool stuff over this!

[Andon Labs @andonlabs]

We've never seen this before.

The biggest jump in Vending-Bench history. GPT-6 Astra is better at making money and more ethical than Claude Fable 5.1.

Surprising, because:
1. First time ever that OpenAI is #1 on Vending-Bench
2. The best model is no longer the unethical one.

[Sam Altman @sama]

Images 2.5 is here.

I don't think it can solve super difficult math problems, but it is really good and we hope you enjoy it.

https://t.co/LdZhGF5AcZ

[OpenAI @OpenAI]

We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.

The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.

The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.

[sarvesh kapre @_cyberhusky]

wild to see how well GPT-6 Astra generalizes across all the domains

[sarvesh kapre @_cyberhusky]

GPT-6 Astra image recognition is unmatched.

In Yellowstone on the way to Lamar Valley, saw a bunch of people staring into the trees. Spent like 15 mins trying to figure out what everyone was looking at lol, finally took a photo and asked ChatGPT - it spotted the tiny osprey nest.

just felt like memory + intelligence + visual recognition + spatial reasoning all came together perfectly. kinda insane tbh

[sarvesh kapre @_cyberhusky]

huge improvement in computer vision capabilities

[Roboflow @roboflow]

GPT-6 Astra is now the top model for vision tasks

sure 3-10x more expensive but 50%+ faster

extremely capable model, OpenAI really pushed the frontier in all directions with this one

[sarvesh kapre @_cyberhusky]

GPT-6 Astra in a league of its own

[Gregor Zunic @gregpr07]

Astra just obliterated our hardest benchmark 👀

Browser Use Benchmark v2 scores are

> Astra medium: 77.3%
> Opus 5: 50.5%
> GPT-5.6 Sol xhigh: 49.1%

22/60 tasks got full marks. Opus 5 had zero.

(fable rejects too many)