Luke Melas-Kyriazi (@lukemelas) profile picture

Building @SpaceXAI | Rhodes Scholar, Oxford PhD

San Francisco, CA
Joined October 2020
Photo by Luke Melas-Kyriazi (@lukemelas)Photo by Luke Melas-Kyriazi (@lukemelas)Photo by Luke Melas-Kyriazi (@lukemelas)
Excited to release our new model! It's more intelligent, better at longer-horizon tasks, just as fast, and the same price
Introducing Grok 4.6. It delivers frontier intelligence and is a significant improvement over Grok 4.5 at the same price.
1
1
44
3,168
Luke Melas-Kyriazi retweeted
We've partnered with SpaceXAI to train Grok 4.5. It’s our most powerful model yet and the first we've built for more than software engineering.
764
1,494
824
17,695
5,310,845
Luke Melas-Kyriazi retweeted
Introducing Composer 2.5, our most powerful model yet. It's more intelligent, better at sustained work on long-running tasks, and more reliable at following complex instructions. For the next week, we’re doubling the included usage of the model.
904
1,280
976
13,058
20,295,919
Very excited to release a new economics paper with @SuproteemSarkar! Lots of data here on how improvements in AI capabilities change the way developers work.
We partnered with University of Chicago economist @SuproteemSarkar to study how more capable models have changed the way people use Cursor. Across 500 teams, we find that developers are tackling more ambitious work with AI, with a 68% increase in high-complexity tasks this year.
5
1
27
2,756
Key findings: - On task complexity: Usage initially grows faster in lower-complexity tasks, but then accelerates in higher-complexity tasks - On firm flexibility: Smaller, private, and newer firms had greater increases in usage - On variation across industry: Media, software, and financial firms grew fastest - On task distribution: AI is used relatively more for architecture, documentation, and code review work
1
7
759
The full paper is available at suproteem.cc/intelligence Hope you enjoy it and feedback is welcome!
1
6
409
It's been clear for a while that long-running cloud agents are the future of coding, but I rarely used them because they were a pain to set up and work with. My favorite part of our new interface is that is makes it seamless to switch between local and cloud!
We’re introducing Cursor 3. It is simpler, more powerful, and built for a world where all code is written by agents, while keeping the depth of a development environment.
6
40
2,596
We train in prod and ship a new checkpoint to users every five hours!
Earlier this week, we published our technical report on Composer 2. We're sharing additional research on how we train new checkpoints. With real-time RL, we can ship improved versions of the model every five hours.
3
3
116
7,726
Lots of alpha in here, especially on RL science, systems, and infra. Enjoy!
We're releasing a technical report describing how Composer 2 was trained.
2
6
102
8,352
Our first frontier-level model! It's the result of our first continued pretraining run as well as further scaling RL. Very excited to hear how people like it! Feel free to send me feedback and we'll incorporate it into future models.
Composer 2 is now available in Cursor.
6
3
90
5,433
Go @JasonBud and @ryolu_! Excited to close the loop on design and engineering with Cursor
You can now design directly in your codebase. Select elements, modify them visually, and Cursor writes the code.
12
2,236
Luke Melas-Kyriazi retweeted
We're super excited to host our third NeurIPS reading group with @LukeMelas, including lunch sponsored by @BoxGroup! We'll gather 25-30 researchers to discuss Composer, @Cursor_AI's frontier agent model trained with RL, RSVP below!
1
6
1
12
3,538
Would highly recommend checking out @SuproteemSarkar's paper, it's full of insights about how developers use agents and how much more productive they are becoming. "A difference-in-differences analysis of weekly code merges indicates that software output increased by 39% after the agent became [Cursor]'s default code generation mode."
Who uses AI agents? How do agents impact output? How might agents change work patterns? New working paper studies usage + impacts of coding agents (1/n)
1
4
1,954
Very excited to release our first agent model!
Introducing Cursor 2.0. Our first coding model and the best way to code with agents.
4
2
33
1,946