Will GPT-4-class input token prices decline by more than 70% YoY by Q4 2026?

Current AI model deployment and development costs remain high, with significant investments in infrastructure, including GPUs. As of Q3 2023, input token prices have been stable, and the demand for advanced AI capabilities continues to grow, making a 70% decline by Q4 2026 unlikely. Historical trends show that while prices for technology tend to decline, such a steep drop is rare unless there’s a radical shift in technology or production efficiency that significantly reduces costs.

The rapid pace of AI hardware innovation and increased competition among major AI providers (OpenAI, Google, Anthropic, Microsoft) are strong drivers for cost reduction. Historically, nascent technologies like GPUs and cloud computing have seen significant price drops after initial adoption, with input token costs likely to follow a similar trajectory as economies of scale and efficiency improvements are realized. However, the current high demand and complexity of training and running GPT-4 class models introduce significant cost pressures that could temper the rate of decline.

Historical precedent shows 40-50% price declines over similar timeframes, with the incremental 20+ percentage points to reach 70% requiring exceptional competitive pressure or strategic pricing changes by late 2026. While infrastructure costs decline steadily and competition intensifies, maintaining >70% YoY decline from Q4

GPT-4-class input pricing fell from $30/M tokens (Mar 2023) to $2.50/M (GPT-4o, May 2024), an 8× or 87.5% decline in 14 months; training runs at 5–6× prior FLOP scale now occur every 10–12 months while inference hardware efficiency improves ~2.5×/year (NVIDIA H100→B200 roadmap). Sustained capex at $50–70 B/yr by Microsoft/OpenAI and continued 30–40% gross-margin targets imply further 3–4× cost cuts are required by late 2026, making an additional 70% YoY drop from the 2025 baseline likely.