AI usage has long carried costs, though average consumers often don't notice due to free entry-level chatbots. Intensive users now pay per token, a model pioneered by Chat GPT in 2022 and adopted by others. Token pricing varies by model sophistication: Anthropic's Claude family includes Haiku (cheapest), Sonnet, Opus, and Fable (most expensive due to greater reasoning capability).
Input tokens are cheaper than output tokens, which require more computation. As AI agent deployment increases, token consumption and bills have risen sharply. Early AI providers offered free access to boost adoption, similar to Uber's early strategy, but surging training and infrastructure costs forced monetization.
Despite growing capacity, aggregate token prices have fallen, with the LLM Token Expenditure Index dropping to 95 US cents in early October — its lowest in nine months. Low switching costs, due to infrastructure-agnostic models, drive strategies like Google bundling cloud storage with Gemini subscriptions to create stickiness. Companies must now manage token usage and measure AI value as pricing evolves.
Source: Financial Times Companies · Summarized by HeadlinesBriefing