Act on this, mostly facts

Opus 5.5 planner with three local workers cost 2 dollars

What to do

Try Opus as planner over cheaper local workers if you want to cut spend on multi-model runs.

Why

One Reddit run measured Opus spend of 41 to 95 cents per worker model, but it is a single self-disclosed test.

What the research found

A single-run Reddit test with Opus 5.5 as planner over three local workers found Opus spend of $0.41 to $0.95 depending on worker model, with Opus reading 48-76x more than it wrote; one run per model, self-disclosed author of the tool.

Claims and how well the evidence supports them

Opus 5.5 spend ranged from $0.41 to $0.95 depending on which local worker model was used 99%
Opus read 48-76 times more tokens than it wrote 97%
Nemotron 3.5 Lightning edits failed about 83 percent of the time (23 ok vs 109 failed) 99%

Evidence

reddit.com https://reddit.com/r/ClaudeAI/comments/1xczls1/opus_55_orchestrated_three_local_models_and_spent

Actionability rated medium.