1 comments

  • mahadansar an hour ago

    Built this mostly to see if parallel workers would actually beat one strong Codex agent once tasks got bigger.

    So far they haven’t. On the largest benchmark Sol had 6 independent modules it could delegate, but chose 0 workers. Forcing 6 workers was around 2x slower.

    Repo has the raw results + benchmark setup if anyone wants to poke holes in it. I’d be curious if people see a different result on much bigger real-world repos.