AI · 1h ago
Gemini 3.6 Flash's Thinking Dial Cuts Costs 30x
Google's Gemini 3.6 Flash charges for reasoning tokens, which can be controlled via a per-request dial. On a 120-word task, the default setting cost $0.03316 while minimal effort cost $0.00110, a 30x difference. However, minimal effort breaks multi-step math tasks, dropping accuracy from 3/3 to 0/3.
Meridian48 take
The cost savings are real, but the trade-off in reasoning quality means developers must carefully choose when to dial down thinking effort.
Read the full reporting
Gemini 3.6 Flash: the Thinking Dial That Moves Cost 30x (Measured) →
DEV Community
gemini-3-6-flashai-cost-optimization