The New Cost-First Debate in Enterprise AI
The growing shift from Anthropic models to cheaper AI model alternatives is a pricing-driven trend in which startups replace premium, closed-source systems with lower-cost or open models to protect margins and extend runway while keeping acceptable quality for production workloads. This trend is no longer quiet. Microsoft AI CEO Mustafa Suleyman has said that “Anthropic is extremely expensive” and that many customers are now looking for alternatives, bringing Anthropic pricing costs into the spotlight for enterprises and startups alike. His remark that Microsoft’s goal is to “reduce and ultimately eliminate” what it pays Anthropic signals how strategic and significant these costs have become, even for a tech giant. For smaller, high-volume AI products, inference is often the largest line item, so any price gap between models compounds quickly at scale.
Lindy’s Full Switch: DeepSeek vs Anthropic in Practice
The clearest current example of DeepSeek vs Anthropic comes from Lindy, an AI agent platform whose product continuously runs models for users. Founder and CEO Flo Crivello decided to “pull the trigger” and move 100% of Lindy’s traffic from Anthropic models to DeepSeek V4. He reports that the company now saves millions of dollars while seeing an increase in performance on many core use cases, turning what might have been a risky migration into a business upgrade. For Lindy, inference had been its number one cost, even larger than payroll, so a 2–5x cut in that line item was described as transformative. The switch required months of benchmarking and far more engineering effort than expected, yet the resulting enterprise AI savings show why pricing can outweigh the friction of migration when volumes are high.
Why DeepSeek’s Pricing Model Changes the ROI Math
DeepSeek’s appeal lies in how it reshapes return on investment for high-traffic AI products. DeepSeek V4-Pro is priced at USD 3.48 (approx. RM16.00) per million output tokens, which makes it far cheaper than many frontier closed-source models. On the Artificial Analysis Intelligence Index benchmark, running V4-Pro costs USD 1,071 (approx. RM4,900.00), compared with USD 4,811 (approx. RM22,000.00) for Claude Opus 4.7, more than a fourfold difference on the same workload. For startups invoking models billions of times per month, that gap converts directly into substantial enterprise AI savings. What makes this more disruptive is that Lindy reports performance gains on critical tasks after switching, especially on agentic, real-world workloads where V4-Pro has scored strongly. Cost is no longer framed as a trade-off against quality but as a path to better economics and, in some cases, better output.
Open Models, Experimentation, and the Future of Anthropic Pricing
For smaller companies, Anthropic pricing costs are encouraging aggressive experimentation with open models and newer providers. Crivello’s team evaluated multiple options, including models like GLM and Kimi K2.5, before deciding that DeepSeek V4 was “way way better” for their use cases. They also selected Atlas Cloud as their hosting provider after an extensive search, highlighting how an ecosystem of alternative infra and models is forming outside the traditional big-lab stack. Importantly, Anthropic is not entirely gone from Lindy’s stack; Claude is still used internally under a generous maximum plan subsidy and reserved for edge cases when Lindy’s default stack fails. This reflects a broader pattern: Anthropic remains a popular enterprise choice, but startups under budget pressure are carving out core workloads for cheaper AI model alternatives, keeping premium models in a narrower, high-value role.
