What the Anthropic Pricing Backlash Is Really About
The shift from premium AI providers to open source AI alternatives is a cost-driven move by startups and platforms replacing proprietary models with cheaper, nearly comparable open systems to protect margins, scale usage, and avoid overdependence on a single vendor’s pricing power. This backlash crystallised when Mustafa Suleyman, Microsoft’s AI CEO, described Anthropic as “extremely expensive” and said “many people are urgently looking for alternatives,” adding that Microsoft’s goal is to “reduce and ultimately eliminate” what it pays Anthropic. At the same time, startups that run large volumes of model calls report that anthropic model cost has turned inference into their dominant expense line. As usage grows into billions of tokens, even modest per-token price gaps compound into painful AI cost savings opportunities that are too large to ignore, especially for early-stage companies under pressure to reach profitability.
Lindy’s DeepSeek Switch: Millions Saved and Better Performance
The clearest signal of this change came from Lindy, an AI agent platform whose product depends on continuous inference. Founder Flo Crivello announced that Lindy had moved 100% of its traffic from Anthropic to DeepSeek V4, saying the switch “saves us millions of $ and we’re actually seeing an increase in performance on many core use cases.” For a company where inference spend had been “#1 cost by a lot (more than payroll),” cutting that line item by 2–5x was described as transformative. DeepSeek V4-Pro is priced at USD 3.48 (approx. RM16.06) per million output tokens and, on the Artificial Analysis Intelligence Index, costs USD 1,071 (approx. RM4,943) to run the full benchmark compared with USD 4,811 (approx. RM22,209) for Claude Opus 4.7. In a deepseek vs anthropic comparison at Lindy’s scale, that difference directly translates into large, recurring AI cost savings.
How Open Models Are Catching Up on Capability
Lower prices alone would not be enough if open source AI alternatives were clearly weaker. What makes DeepSeek and peers compelling is that the capability gap has narrowed for many production workloads. DeepSeek V4 launched with two variants, V4-Pro and V4-Flash, and V4-Pro scored 1554 on GDPval-AA, an agentic real-world tasks benchmark from Artificial Analysis. DeepSeek itself says it trails the US frontier by about 3–6 months, but for agent-style work such as scheduling, summarisation, and structured decision support, buyers like Lindy say that difference no longer matters. In practice, this puts DeepSeek in direct competition with enterprise AI pricing from leading proprietary labs. For teams optimising deepseek vs anthropic trade-offs, the question is less “Which is absolutely best?” and more “Which model is good enough at a fraction of the cost for our specific tasks?”
Enterprise AI Pricing vs Startup Budget Reality
Anthropic’s strongest position remains in large, well-funded enterprises that value support, compliance, and deep integrations, and can absorb higher anthropic model cost within broader digital transformation budgets. Flo Crivello expects Anthropic to “be fine because of enterprise relations,” and Lindy still keeps Claude available internally and for edge cases where their default stack fails. This highlights a clear market split. One segment will keep paying for premium enterprise AI pricing, optimising for reliability, vendor relationships, and perceived frontier status. Another segment, dominated by startups and cost-conscious companies, is standardising on open or open-weights models, often hosted by third parties like Atlas Cloud, to gain price control and flexibility. As Mustafa Suleyman frames it, even a giant buyer like Microsoft is working to remove external model spend, which underscores how exposed smaller companies feel when a single AI provider can dictate the economic structure of their products.
What This Shift Signals for the AI Model Market
The move from Anthropic to DeepSeek is not a one-off stunt; it previews a broader recalibration of how companies buy AI. For many teams, AI cost savings now sit beside accuracy and latency as a core dimension in architecture decisions. When a leading startup can churn from a flagship proprietary model, endure a migration described as “100x more work than we thought,” and still see business upside, it encourages others to follow. For incumbents, this likely means sharper product tiering: frontier models for the most demanding enterprise AI use cases, and cheaper families for high-volume workloads. For open providers, it is validation that they can serve not only hobbyists but also serious commercial deployments. The next phase of competition will turn on who can pair strong performance with sustainable, transparent pricing that startups can build a business on.
