Report
Self-modified coding-agent harness reportedly matches Codex in a nine-harness comparison
A post sharing SelfSearch research says the agent solved 82.0% of Terminal-Bench 2.1 with DeepSeek V4 Flash after a $4.03 search.
TLDR
A post sharing SelfSearch research says a coding agent rewrote its own instructions, tools and procedures using records of earlier attempts, without task rewards during the search. With DeepSeek V4 Flash, the modified harness solved 82.0% of Terminal-Bench 2.1, matching Codex in a public nine-harness comparison under the same settings. The search cost $4.03.
Combined views
1 Source, first seen ago
