Naman Jain @StringChaos

Research @cursor_ai | CursorBench, LiveCodeBench, DeepSWE, R2E-Gym, GSO, LMArena Coding | Past: @UCBerkeley @MetaAI @AWS @MSFTResearch @iitbombay naman-ntc.github.io San Francisco, CA Joined March 2018

Tweets

565
Followers

3K
Following

1K
Likes

5K

Naman Jain @StringChaos

2 weeks ago

@novasarc01 Thats why we built CursorBench! cursor.com/blog/cursorben…

0 0 5 493 3

View Details

elie @eliebakouch

3 weeks ago

correlation between CursorBench and Artificial Analysis reported scores benchmarks like IFBench or tau2 show ~0 correlation with CursorBench. opus 4.7 (max effort) performs relatively better on CursorBench than on other benchmarks, gpt 5.5 shows the opposite pattern

Michael Truell @mntruell

3 weeks ago

Gemini Flash 3.5 is now on CursorBench, our main coding agent eval. We’ll keep updating the leaderboard as new models come out. cursor.com/evals

108 88 1K 1.5M 225

10 7 158 25K 55

View Details

Michael Truell @mntruell

3 weeks ago

Gemini Flash 3.5 is now on CursorBench, our main coding agent eval. We’ll keep updating the leaderboard as new models come out. cursor.com/evals

108 88 1K 1.5M 225

View Details

Naman Jain @StringChaos

4 weeks ago

Check out Composer 2.5, our new model pushing pareto frontier

Cursor @cursor_ai

4 weeks ago

Composer 2.5 is exceptionally intelligent and up to 10x more efficient than similarly capable models.

55 115 2K 459K 185

0 0 26 2K 0

View Details

Hao Wang @MogicianTony

2 months ago

SWE-bench Verified and Terminal-Bench—two of the most cited AI benchmarks—can be reward-hacked with simple exploits. Our agent scored 100% on both. It solved 0 tasks. Evaluate the benchmark before it evaluates your agent. If you’re picking models by leaderboard score alone, you’re optimizing for the wrong thing. 🧵

22 90 679 828K 370

View Details

Cursor @cursor_ai

3 months ago

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.

101 129 2K 507K 503

View Details

Naman Jain @StringChaos

3 months ago

@koushik77 For the first probably, almost no model considers the esbuild transpiler issue! For the second, agents can actually tune approximate algorithm quite well.

0 0 1 223 0

View Details

Naman Jain @StringChaos

3 months ago

Check out the tech report detailing our continued pre-training and RL setup behind Composer2! Also sharing some example CursorBench problems by popular demand

Cursor @cursor_ai

3 months ago

We're releasing a technical report describing how Composer 2 was trained.

169 479 5K 1.3M 4K

1 3 48 3K 4

View Details

Sasha Rush @srush_nlp

3 months ago

It's really neat to see all the interest in the Composer 2 technical report, from training to kernel design to inference. If you have any questions about why we did things, feel free to ask. I'll run around the office and bug people.

Cursor @cursor_ai

3 months ago

We're releasing a technical report describing how Composer 2 was trained.

169 479 5K 1.3M 4K

35 18 320 58K 103

View Details

Naman Jain @StringChaos

3 months ago

Excited to share Composer-2 with everyone. It has come a long way since Composer-1, still lots more to go! Hope you like it!

Cursor @cursor_ai

3 months ago

Composer 2 is now available in Cursor.

647 879 10K 5.4M 2K

3 5 66 3K 1

View Details

Cursor @cursor_ai

3 months ago

We trained Composer to self-summarize through RL instead of a prompt. This reduces the error from compaction by 50% and allows Composer to succeed on challenging coding tasks requiring hundreds of actions.

91 96 2K 229K 375

View Details

Naman Jain @StringChaos

3 months ago

Check out full post at: cursor.com/blog/cursorben…

0 1 10 910 9

View Details

Naman Jain @StringChaos

3 months ago

Lots more details in the post: 1. Pareto frontier across different metrics 2. How CursorBench has shifted as agent capabilities changed 3. CursorBench vs public evals: what’s missing and future work directions 4. CursorBench vs online: how online metrics shape offline evals

1 1 12 1K 1

View Details

Naman Jain @StringChaos

3 months ago

New post: how we do evals at @cursor_ai. Takeaways: 1. Online metrics from real Cursor requests provide construct validity 2. CursorBench: a dynamic offline suite distilled from online learnings 3. Multi-axes evals -- correctness, efficiency, agent interaction behavior

Cursor @cursor_ai

3 months ago

We're sharing a new method for scoring models on agentic coding tasks. Here's how models in Cursor compare on intelligence and efficiency:

210 253 3K 633K 922

5 18 147 39K 103

View Details

Manish Shetty @slimshetty_

3 months ago

GSO Update. gpt-5.4 (xhigh) scores 31.4% with reasoning_effort=high, gpt-5.4 slightly lower than gpt-5.2. a quick thought on why below:

3 4 61 7K 6

View Details

Manish Shetty @slimshetty_

4 months ago

x.com/i/article/2023…

2 6 54 24K 41

View Details

Cursor @cursor_ai

4 months ago

Long-running agents are now available at cursor.com/agents for Ultra, Teams, and Enterprise plans. With our new harness, agents can complete much larger tasks. cursor.com/blog/long-runn…

61 95 971 365K 295

View Details

Cursor @cursor_ai

4 months ago

Composer 1.5 is now available. We’ve found it to strike a strong balance between intelligence and speed.

155 183 2K 665K 242

View Details

Michael Truell @mntruell

5 months ago

We built a browser with GPT-5.2 in Cursor. It ran uninterrupted for one week. It's 3M+ lines of code across thousands of files. The rendering engine is from-scratch in Rust with HTML parsing, CSS cascade, layout, text shaping, paint, and a custom JS VM. It *kind of* works! It still has issues and is of course very far from Webkit/Chromium parity, but we were astonished that simple websites render quickly and largely correctly.