AI token costs fall as enterprises prioritize ROI over price

The cost of AI inference is declining rapidly, with the price of processing one million tokens dropping approximately tenfold annually. While lower costs suggest increased usage, enterprise focus is shifting from the price of tokens to their measurable return on investment and productivity gains.

A comparison of deployable models reveals significant price and performance disparities. DeepSeek V4 Flash offers a cost of $0.15 per million input tokens and $0.29 per million output tokens, whereas Meta’s Muse Spark 1.2 is priced at $1.25 per million input tokens and $4.25 per million output tokens. On Artificial Analysis’ cost-per-task measurement, Muse Spark 1.2 is roughly 13 times more expensive than DeepSeek V4 Flash.

Performance metrics also differ sharply. DeepSeek V4 Flash demonstrates significantly lower latency, with a median first-token latency of 444 milliseconds, compared to 7.73 seconds for Muse Spark 1.2. While Muse Spark 1.2 holds a slight edge in certain intelligence indices, the economic and speed advantages favor cheaper, open-weight models like DeepSeek for many production environments.