Lindy's 25 Staff Switched 100% Traffic from Anthropic to DeepSeek with Zero User Complaints During Outage

This isn't simply about being "cheap," but rather a mathematical domination in cost-performance on the inference side. I ran the numbers: DeepSeek-V2's API pricing is ¥0.14 per million input tokens and ¥0.28 per million output tokens. Compared to Anthropic Claude 3 Opus at $15 per million input tokens and $75 per million output tokens, converted at current exchange rates, DeepSeek's price is roughly 1/10th to 1/20th of Anthropic's. However, if Lindy, an AI assistant for end-users, saw a noticeable drop in model quality, users would immediately perceive it and churn. The lack of complaints suggests that DeepSeek performs at least at an "acceptable" or even "good" level in most conversational scenarios.