Rishi Desai

Rishi Desai researches coding agents at Meta Superintelligence Labs. At Abundant AI he built RL environments for agentic coding and created SWE-Marathon, a benchmark of 20 project-scale software tasks where frontier agents solve under 30%.

Career

Cosigned byLoading...

CosignedSign in to view

Elsewhere

Abundant

Until Oct 2026

Machine Learning Engineer

Stanford University

BS and MS, Computer Science

Products & media

arxiv.org

SWE-Marathon: Can Agents Autonomously Complete Ultra-Long-Horizon Software Work?

First-author paper introducing a 20-task benchmark for long-horizon software work; agents average 27M tokens per attempt and solve under 30% of tasks.

Writing

youtube.com

SWE-Marathon: Evaluating Coding Agents at Billion-Token Scale

AI Engineer World's Fair 2026 talk on benchmarking coding agents on whole projects, including reward hacking.

Video

Profile name

Company name

Company name

Profile name

Company name

Company name

Profile name

Company name

Company name

Profile name

Company name

Company name

Profile name

Company name

Company name

Rishi Desai

Meta

Abundant

Stanford University

Products & media

SWE-Marathon: Can Agents Autonomously Complete Ultra-Long-Horizon Software Work?

SWE-Marathon: Evaluating Coding Agents at Billion-Token Scale

Curated Lists

List Inclusions