rajan agarwal

@_rajanagarwal

research @openai

rajan.sh

Joined March 2018

Replies Reposts Media

Pinned
rajan agarwal
@_rajanagarwal
Nov 25, 2025
in order to have research agents that can run for days, we need context compaction i used RL to have LLMs naturally learn their own 10x compression! Qwen learned to pack more info per token (ie use Mandarin tokens, prune text) read the technical blog:
rajan.sh
LLMs can invent their own compression - Rajan Agarwal
As a constrained optimization problem, LLMs can use RL to invent their own compression schemes to increase its context window.
28
rajan agarwal
@_rajanagarwal
Sep 22
welcome sol and luna! major improvements in pareto intelligence and taste
OpenAI
@OpenAI
Sep 22
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
00:00
3
rajan agarwal
@_rajanagarwal
Sep 18
ill be judging @HackTheNorth & doing a couple talks about research and codex as an @OpenAI sponsor! come find me at the booth / pop by the tech talks/workshops
8
rajan agarwal
@_rajanagarwal
Sep 17
great to see Astra’s multimodal and long horizon capabilities!
Proximal
@ProximalHQ
Sep 17
GPT-6 Astra is the best-performing model on FrontierSWE Astra achieves a score of 65.5%, outperforming Fable 5.1 (56.3%) and its predecessor GPT-5.6 Sol (32.2%)
rajan agarwal
@_rajanagarwal
Sep 16
super cool for this to be public
Fuli Luo
@_LuoFuli
Sep 16
Nearly half a year of silence. We spent it studying one problem: how far RL can scale. MiMo-V2.6 is in the middle of its RL run right now. Three things we scaled: compute (~2B tokens per step, 1568 prompts × 16 rollouts, fully async), environments and harnesses (multi-task
3