Will AI token prices keep falling?

The curve is long-run. Inference for a fixed level of capability dropped from $60 per million tokens in 2021 to $0.06 in 2024 (a16z). But the binding constraint keeps migrating, from chips to power to memory to data, so short-term rents on scarce hardware still rise. Nor is the pressure even: it concentrates on the most substitutable cheap tier, while the flagship tier has no substitute, and pricing power is set by whether a tier can be swapped out. Forecasting cost from one average token price will misstate it.

On July 30, 2026 OpenAI cut GPT-5.6 Luna by 80%, from $1 to $0.20 per million input tokens and $6 to $1.20 output, while flagship Sol held at $5 input and $30 output, three weeks after the family launched on July 9 (confirmed).

Budget against the curve, never against one vendor's price list. FutureX's practice: measure spend as capability per dollar, keep two providers live, let batch work chase the cheapest open weights, and tie the AI line to gross margin so it grows with what it earns. The third of FutureX's three screens for application companies is whether gross margin climbs as token prices fall; a business model whose margin assumes upstream prices only ever fall needs to re-run that stress test.

From August 16, 2026 DeepSeek moved V4-Pro to peak and off-peak pricing; peak output rose from $0.87 to $3.96 per million tokens, as much as 1,100% across tiers, with off-peak at half price (confirmed). On August 23 Gavin Baker reported that rents on multi-thousand-GPU B200 clusters rose 50 to 60% in seven months (confirmed).