/Tech1h ago

Recursive, co-founded by Cong Lu, launches automated AI system that optimizes nanoGPT and NVIDIA GPU kernels

Story Overview

Recursive has released its first public results from an automated AI research system that independently discovers optimizations for model training pipelines and GPU kernels, posting state-of-the-art numbers on nanoGPT and NVIDIA benchmarks while open-sourcing every artifact produced.

119319264.6K
Original post
Cong Lu@cong_ml#1693inTech

Recursive just came out of stealth, and the team has been cooking 🔥

Our first results: an automated AI research system that can improve AI across 3 very different settings across training and GPU kernel optimization.

https://www.recursive.com/articles/first-steps-toward-automated-ai-research

1:47 AM · Jun 11, 2026 · 3K Views
Developer Impact

Open-sourced outputs lower the barrier for reuse

Anyone can now grab the exact optimizations the system found for nanoGPT training and NVIDIA kernels instead of repeating the search themselves.

Open Question

Scale of the next leap stays unspecified

The three benchmark settings show early capability, yet no information is given on compute required, architecture details, or how far the approach extends beyond these tasks.

Sentiment

Users are excited about Recursive's autonomous AI research system because its automated loop enables cross-domain progress and new SOTAs on NanoGPT and NVIDIA kernels.

Pos
100.0%
Neg
0.0%
9 comments with sentiment.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS463BOOKMARKS2RETWEETS3REPLIES1
Tim Shi@timshi_ai

Recursive’s autonomous research system just set new SOTAs on @karpathy’s nanogpt + nanochat and @nvidia’s kernel benchmarks!!

No humans in the loop. The system discovered these gains on its own.

Open-sourcing results today.

Recursive@Recursive_SI

http://x.com/i/article/2064569799313940481

42mViews 463Likes 6Bookmarks 2
LIKES10
Cong Lu@cong_ml

Domain 2: squeezing more out of a mature highly optimized training benchmark ⚡

On NanoGPT Speedrun, starting from a solution refined by the community over 2+ years, the system still found a sequence of discoveries to produce a further speedup: 79.7s → 77.5s.

Cong Lu@cong_ml

Domain 1: fixed-budget LM training ⏱️

On NanoChat Autoresearch, our system found a training recipe that reaches the same loss 1.3x faster than the best community solution, and substantially improves over the initial hand-optimized baseline.

1hViews 236Likes 10Bookmarks 1
Cong Lu@cong_ml

Domain 1: fixed-budget LM training ⏱️

On NanoChat Autoresearch, our system found a training recipe that reaches the same loss 1.3x faster than the best community solution, and substantially improves over the initial hand-optimized baseline.

Cong Lu@cong_ml

Recursive just came out of stealth, and the team has been cooking 🔥

Our first results: an automated AI research system that can improve AI across 3 very different settings across training and GPU kernel optimization.

https://www.recursive.com/articles/first-steps-toward-automated-ai-research

1hViews 350Likes 7Bookmarks 1
Tim Shi@timshi_ai

Towards a system that discovers knowledge autonomously!

New SOTAs on @karpathy’s nanogpt & nanochat

and @nvidia’s kernel benchmarks.

We are open-sourcing the artifacts of the system and contributing back to the research community!

Recursive@Recursive_SI

http://x.com/i/article/2064569799313940481

1hViews 404Likes 6Bookmarks 0
Alok Bishoyi@alokbishoyi97

@cong_ml pretty cool stuff @cong_ml ! have been hacking away at autoR loops / harness setups myself as well , all opensourced at @evo__hq

44mViews 83Likes 3Bookmarks 1
Cong Lu@cong_ml

Domain 3: low-level GPU kernel optimization ⚙️

On Nvidia’s SOL-ExecBench, the same general system improved mean SOL score from 0.699 to 0.754 across 235 kernels - an 18% reduction in the gap to the theoretical optimum.

Cong Lu@cong_ml

Domain 2: squeezing more out of a mature highly optimized training benchmark ⚡

On NanoGPT Speedrun, starting from a solution refined by the community over 2+ years, the system still found a sequence of discoveries to produce a further speedup: 79.7s → 77.5s.

1hViews 127Likes 6Bookmarks 0
Tankred Saanum@TankredSaanum

We're releasing our blog post along with a lot of the artifacts our system came up with optimizing these problems. Lots of interesting techniques and implementations to dig into: https://github.com/recursive-org/first-steps-toward-automated-ai-research

1hViews 24Likes 4
Kamesh 🇺🇸@ElangovanKamesh

@cong_ml The value is no longer in a single breakthrough but in a system that can keep generating them.

22mViews 15Likes 2
Cong Lu@cong_ml

What’s exciting to me is not just any one result.

It’s that one system can make progress across problems with very different bottlenecks: model quality under a compute budget, wall-clock training speed, and hardware-level kernel performance.

1hViews 11
Tankred Saanum@TankredSaanum

Thrilled to finally share some of the work we at @Recursive_SI has been doing since we launched!🔥

We've made a system that autonomously conducts AI research and tested it across three different settings: Small Language Model training, NanoGPT speedrunning, and kernel engineering!

A bit like Karpathy's Autoresearch, but scaled up and designed to be open-ended, we can push our system to optimize models, training algorithms, and kernels in really cool ways.

Blogpost: https://www.recursive.com/articles/first-steps-toward-automated-ai-research

1hViews 465Likes 11Bookmarks 2
Tankred Saanum@TankredSaanum

Also super cool to see that the system works across several domains, not just Language model training. Super excited to keep pushing with this fantastic team!

1hViews 17Likes 3
Tim Shi@timshi_ai

Towards a system that discovers knowledge autonomously!

New SOTAs on @karpathy’s nanogpt + nanochat and @nvidia’s kernel benchmarks.

We are open-sourcing the artifacts of the system and contributing back to the research community!

Recursive@Recursive_SI

http://x.com/i/article/2064569799313940481

47mViews 57Likes 1Bookmarks 0
Cong Lu@cong_ml

@sharathraparthy Thank you 🫶

1hViews 43Likes 1
未知@luyun0120

@cong_ml AI现在最大的问题是:说得比做的好听。真正能落地的项目,远比融资PPT里的少。 「Recursive just came out of stealth, and the team has been co...」

1hViews 33Likes 1
lovish@louvishh

initial results on automated ai research from our team on small scale pre-training and kernel optimization. we also open-source the corresponding artifacts.

it's been great seeing all the amazing progress here in such a short time!

Recursive@Recursive_SI

http://x.com/i/article/2064569799313940481

8mViews 18Likes 2Bookmarks 0
Cong Lu@cong_ml

@ElangovanKamesh Endless compounding discovery!

22mViews 12
Cong Lu@cong_ml

The common thread is the automated research loop.

Search over ideas, turn them into executable changes, test them, validate + cross-check the result, and use empirical evidence to plan the next ideas.

Extremely excited to keep pushing @Recursive_SI !!

1hViews 10
Loewen Cavill@loewenkc

@timshi_ai @karpathy @nvidia Damn such fast progress. Crushing it.

33mViews 4
Sharath Raparthy@sharathraparthy

@cong_ml Congrats! Cool kernel optimisation results 😃

1hViews 1