The Substitution Wave in AI

2026-06-07 · Source: Tomasz Tunguz · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Intermediate, quick

Summary

The substitution wave in AI is driven by three forces: foundation labs moving into applications, rising frontier model prices, and open-source models achieving "good enough" quality for most use cases. AI buyers are responding by substituting expensive frontier models with more cost-effective alternatives. Coinbase routes prompts to cheaper models, keeping costs flat despite exponential token usage. Lindy switched 100% of its traffic from Anthropic models to DeepSeek v4, saving millions of $ and observing increased performance. Harvey's Legal Agent Benchmark showed SFT improved Kimi 2.6's all-pass rate from 11% to 15%, surpassing Opus' 14%, at an 11x lower cost (\$84 vs \$954 for 100 tasks). Cursor post-trained Kimi K2.5 into Composer 2.5, achieving up to 10x more efficiency. Buyers typically reinvest these cost savings into acquiring more intelligence rather than pocketing the discount.

Key takeaway

For Directors of AI/ML evaluating model choices, you should actively explore open-source alternatives like DeepSeek v4 or Kimi K2.5. These models now offer comparable or superior performance to expensive frontier models at a fraction of the cost, as demonstrated by Lindy's millions of $ savings and Harvey's 11x cost reduction. Reinvesting these savings can significantly increase your organization's overall intelligence consumption and capabilities.

Key insights

Open-source AI models now offer competitive performance at significantly lower costs, driving a substitution wave among AI buyers.

Principles

• Open models achieve "good enough" performance.
• Frontier model prices are increasing.
• Cost savings enable higher intelligence consumption.

Method

Companies are routing prompts to cheaper models, switching entire traffic loads, or post-training open-source models to optimize cost and performance.

In practice

• Route prompts to cost-effective models.
• Evaluate DeepSeek v4 for core use cases.
• Post-train Kimi K2.5 for efficiency.

Topics

• AI Cost Optimization
• Open-source Models
• Frontier Models
• Model Substitution
• DeepSeek v4
• Kimi K2.5

Best for: CTO, VP of Engineering/Data, AI Engineer, Director of AI/ML, Consultant, Entrepreneur

Related on AIssential

• China Takes Over — Deepseek V4 offers frontier-level AI performance at a fraction of the cost, challenging established market leaders.
• Open-Source AI Gains Ground as Rising Costs Push Shift to Smaller Models — Open-weight AI models offer substantial cost savings and near-parity performance compared to closed models, yet remain significantly underutilized.
• China is falling behind in the AI race, according to a US government benchmark — Chinese AI models lag US counterparts in capability but offer significant price advantages.
• DeepSeek V4 - almost on the frontier, a fraction of the price — DeepSeek V4 models offer competitive performance and industry-leading low costs through significant efficiency gains.
• New DeepSeek Research - The Future Is Here! — DeepSeek's open research provides a reproducible framework for advanced AI, emphasizing self-optimization and efficient training methods.
• 🔮 X-raying OpenAI’s unit economics — Frontier AI models like GPT-5 face significant profitability challenges due to high R&D costs and short model lifespans.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Tomasz Tunguz.