Opus alone vs Opus planning + local Qwen coding: $2.52 vs $0.62 for the same game
GapNew4766 · reddit · 2026-10-26
A developer ran the same job twice — building a 5-level pixel platformer from one spec on a MacBook Pro M5 Max via Atomic Agent's Fusion mode:
- Opus 5.5 alone: 16 min, $2.52 cloud cost, 90.7k output tokens, 25/25 checklist
- Fusion (Opus plans and reviews, local Qwen 3.8 27B writes code with 3 workers): 42 min, $0.62, Opus output down to 11.3k tokens, 24/25 checklist
Both games work (5 levels, boss, win screen). Since most of the Opus bill is output tokens, offloading the typing to a local model cut cloud costs 75%, at the cost of 2.5x+ runtime because the workers share one GPU. Takeaway: use Opus alone when speed matters; let Opus plan and a local model type for large, splittable jobs.
More from coding & agent
- Dev builds open-source Excalidraw MCP after AI-generated diagrams kept overlapping text and arrows — pikahikmag · 2026-10-28
- DeepSeek vs Haiku pricing debunk: accounting for peak rates and token use, 1017 is 2-2.5x cheaper — teortaxesTex · 2026-10-28
- Viral screenshot shows an agent spawning 32 parallel workspaces in ~30 seconds — cephaloform · 2026-10-28
- Grok Bot 0.78.1 hides unreleased per-bot logins with manager-only team sharing — faraz0x · 2026-10-28
- Uber Built a Legal Redlining Agent for Its Lawyers—Here's What It Learned — AxSaucedo · 2026-10-28
- GraphRAG's hub-node trap: same two-hop Cypher costs 71 DB hits normally, 60,013 on a hub — JeremyCMorgan · 2026-10-28