Same agent build task: DeepSeek costs $0.21 vs Opus 5.5's $2.20
teortaxesTex · x · 2026-10-19
teortaxesTex quotes benjistacked's benchmark: building the same "octopus folding laundry" room, DeepSeek V4.1 Flash 1017 took 22 minutes at $0, while Opus 5.5 took 15 minutes at $2.20 — with nearly identical results.
His own rerun with the same prompt: 21 minutes, 9M tokens (182K output, 508K uncached input), just $0.21 at DeepSeek's off-peak API prices; the screenshot shows context caching saved $1.23 (98%). He concludes DeepSeek is surprisingly close, and the remaining gap is the scene's inherent complexity.
More from coding & agent
- Dev suggests Claude Code compact context before notifying, not after — tonygwu · 2026-10-19
- Consultant open-sources agent-first slide editor where Claude Code edits decks from comments — beh_zod · 2026-10-19
- Chat is the wrong shape for AI brainstorming: a dev experiments with tree-shaped idea maps — S_B_B_B_K · 2026-10-19
- Addy Osmani: writing code isn't the bottleneck anymore — build agent self-verification loops — bibryam · 2026-10-19
- Local models are now good enough for first-round code review, dev finds while paying frontier API prices — draginol · 2026-10-19
- Dev replaces hundreds/month in Modal sandboxes with an idle Mac Mini at home — andersonbcdefg · 2026-10-19