Work

What Is Hard for LLMs, Part 2: Optimization Instead of an Answer (Interim Results)

Year: 2026

Type: report

Abstract: A synthetic benchmark with no ceiling: Claude Code agents on four Claude models make an interpreter for a new virtual machine cheaper; the stronger models score higher but break it more often.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.24011975

Author: Latipov, Aziz

Language: English

Topic: Software Engineering Research

Subfield: Information Systems

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W7225404142

Cites:

Cited by:

Related to: 10