200B Tokens Later: A Month of Letting AI Agents Decompile MW2

I would be interested in comparing this to 5.6 Sol on Codex instead of Claude. There's no moment where I thought claude had beat codex except for about 2 weeks earlier this summer. Things have drastically changed in favor of codex. The harness or something must be better.

I wonder if escalating to Opus or Fable would change the behavior for the larger system level issues. I see the mention of starting with Opus, but now that the project is further along, perhaps a chance to take a second look.

I'd be curious to see how this would run on Qwen3.8 27B