The combination of outsourcing and local AI will soon become more economical than frontier models

While API prices for Frontier models are on an upward trend, local/open-source models like DeepSeek are showing overwhelming price competitiveness.

Recently, major models such as GPT 5.5, Gemini 3.5 Flash, and Anthropic Opus 4.7 have been increasing costs through price hikes or higher token consumption. In contrast, when comparing costs based on agentic tokens, DeepSeek is about 30 times cheaper than Anthropic or OpenAI.

Rather than simply using the highest-performing model, combining 'engineers from low-wage countries + local AI with sufficient performance' can be more efficient. This is because current frontier models excel at task processing but still have limitations in core capabilities needed for engineering, such as long-term memory and autonomous judgment.

Based on simulations of token consumption, engineer salaries, and model price changes, it is predicted that within approximately 11 months, a crossover point will occur where the cost of the 'engineer + DeepSeek' combination becomes lower than the inference cost of frontier models.