The New Economics of AI: Why Cost Now Decides the Model
The shift from premium AI providers to cheaper or open source AI models is driven by a simple reality: AI model costs now dominate many companies’ technology budgets, forcing teams to trade brand-name models for cheaper options that still meet performance needs. Anthropic pricing expensive has become a flashpoint in this debate, especially as usage scales into billions of calls. For startups and enterprises alike, inference can surpass payroll and infrastructure as the main expense, turning cost-performance ratios into board-level decisions. This is reshaping AI adoption strategies: instead of defaulting to the most famous closed models, companies are benchmarking alternatives like DeepSeek to achieve enterprise AI savings while keeping accuracy and reliability. In this new phase, the question is less “What is the best model?” and more “What is good enough at a price we can sustain?”.
Anthropic’s Pricing Squeeze and Microsoft’s Public Pushback
Anthropic’s strongest advocates once praised its safety and quality; now many complain its models have become too expensive for cost-conscious teams. The pressure surfaced openly when Microsoft AI CEO Mustafa Suleyman told Bloomberg that “Anthropic is extremely expensive, and I think many people are urgently looking for alternatives.” He added that Microsoft pays a lot to Anthropic and wants to “reduce and ultimately eliminate that cost,” a striking admission from a major enterprise buyer. The comment signals more than annoyance at a vendor bill: Microsoft aims to be one of the top four labs, building its own frontier models instead of relying on partners. As enterprises hear a major platform provider question Anthropic pricing expensive in public, they gain cover to reassess their own contracts and explore a DeepSeek alternative or other open models for large-scale workloads.
Lindy’s DeepSeek Switch: Millions Saved and No Performance Regret
The abstract debate over AI model costs turned concrete when Flo Crivello, CEO of AI agent platform Lindy, announced his company had dropped Anthropic for DeepSeek V4. He wrote that Lindy “switched 100% of Lindy traffic to DeepSeek v4, churning from Anthropic models. Saves us millions of $ and we’re seeing an increase in performance on many core use cases.” For Lindy, inference was the number-one cost, exceeding payroll, so cutting it by 2–5x promised to transform the business. DeepSeek V4-Pro, at USD 3.48 (approx. RM16.00) per million output tokens, undercuts many closed models. On the Artificial Analysis Intelligence Index benchmark, V4-Pro costs USD 1,071 (approx. RM4,920) versus USD 4,811 (approx. RM22,100) for Claude Opus 4.7, more than four times cheaper. That cost gap, combined with strong agentic performance, made DeepSeek a compelling DeepSeek alternative for Lindy’s production workloads.
Open Source AI Models Mature into Enterprise-Ready Options
Lindy’s move did not happen overnight. Crivello described the migration as “100x more work than we thought,” involving new infrastructure, internal tooling and extensive benchmarking of multiple open models. DeepSeek V4-Pro’s score of 1554 on the GDPval-AA benchmark positioned it as a leading open-weights model for agentic, real-world tasks, the exact profile needed for AI employees that handle complex workflows. DeepSeek itself says it trails the highest-end US frontier models by 3–6 months, but for many production cases that temporal gap no longer matters. The evaluation also included other open and alternative models like GLM and Kimi K2.5, which Crivello called “incredible,” before Lindy settled on DeepSeek. This level of due diligence shows that open source AI models and newer providers can meet demanding enterprise standards when cost, latency and accuracy are measured side by side against closed incumbents.
How Cost-Performance Tradeoffs Are Redrawing AI Strategy
The Lindy example captures a wider shift: AI buyers no longer treat top-tier closed models as default choices. As usage grows and Anthropic pricing expensive remains high, enterprises break their workloads into tiers, reserving the priciest models for narrow, high-stakes edge cases and routing everything else to cheaper options. Even Lindy still uses Claude internally where subsidized plans make sense, and may fall back to Claude Opus when an AI agent fails. But this usage is becoming marginal rather than central. For both startups and large platforms such as Microsoft, the goal is clear: reduce dependence on costly providers, pursue enterprise AI savings and adopt a portfolio of models tuned to specific tasks and price points. In this environment, open models and alternatives like DeepSeek are no longer experiments; they are becoming the backbone of many production AI systems.
