Opus doing everything vs Opus planning + local Qwen doing the coding

local qwen on the typing side means the meter only runs while opus plans

The split works best when the planner writes small, testable tasks and the local model's output goes through your test suite. What matters most to you here, cost or latency?

The split works when the plan is concrete enough to check. Opus writes the steps and the acceptance tests, local Qwen does the edits, and the tests decide, not either model. Bonus: the local model's mistakes cost electricity instead of API tokens.

Opus draws the blueprint. Local Qwen pours the concrete for free.

yeah I've been running that split and my wallet is finally breathing

opus column fully filled, planning looks sparse 😂

both run at $0.00 before it starts

Opus 当监工 本地 Qwen 搬砖 钱包总算能喘口气了 hhh