共有:

Claude · API Cost

Lindy Drops Claude Entirely, Switches to DeepSeek Full-Stack

AI startup Lindy has eliminated all Claude usage and moved entirely to DeepSeek, announcing millions of dollars in annual API savings. The cost-vs-quality calculation has shifted decisively.

AI Navigate Editorial·2026.06.27·6 min read

Claude Anthropic high cost discontinued DeepSeek from China 1/10th cost annual $M saved per year
01
The Decision

What Happened

AI startup Lindy has fully deprecated Anthropic's Claude and migrated entirely to DeepSeek, reporting millions of dollars in annual API cost savings. Six months ago, most startups stayed with Claude despite the premium — the quality gap justified the cost. DeepSeek's accuracy improvements and widening price differential have flipped that judgment.

Lindy's core business — an AI agent builder platform for end users — means API costs hit margin directly. The company says this wasn't a quality compromise: "The quality gap closed. Only the cost gap remained."

02

The Numbers

$M+
annual savings (Lindy's own figure)
~1/10
DeepSeek vs Claude API unit cost
100%
of Claude traffic deprecated
03
Decision Framework

How to Evaluate Your Own API Mix

01

Instrument your API spend by model, monthly

Most companies lack visibility into which models cost how much. Build a monthly dashboard: team × model × spend. You can't optimize what you can't see.

02

Run per-task evals before switching

Not every task needs the top model. Summarization, classification, and scripted responses often work fine on cheaper models. Run blind quality evals per use-case before making a switch decision.

03

Check geopolitical and compliance constraints with legal

Depending on your industry, switching to a China-origin model may carry regulatory, contractual, or data residency implications. Financial, healthcare, and government-adjacent verticals should get legal sign-off first.