AnalysisAI ModelsNovember 3, 2026

Test: cheap coding subagent made Opus 5.5 worse and pricier

Read original source →reddit.com

Across 12 hard coding tasks, Opus alone solved 81% at $1.20/task; pairing it with DeepSeek V4.1 Flash dropped accuracy to 67% while cost rose to $1.29. None of 7 team designs (41 setups, ~1,200 runs) beat Opus alone.

1 source

More stories today

Open the live feed