The cost of running AI is falling. That’s the directional consensus from AI industry tracking in early 2026, and the implications for enterprise procurement are real whether or not the precise numbers hold up to scrutiny.
OpenAI’s current API pricing page reflects updated rates for GPT-4 and other models. AI industry tracking by AI Empire Media reports a reduction of approximately 60% for GPT-4 API access, a figure that has not been independently verified against OpenAI’s official pricing history, and should be treated as reported rather than confirmed. Anthropic’s published pricing shows Claude Haiku 3.5 at $0.80 per million tokens; a specific reduction percentage for Claude 3.5 Sonnet could not be confirmed at the time of publication.
The aggregate claim is larger. According to AI Empire Media, API costs across major providers have dropped 40 to 70 percent in 2026. That range comes from a single source and hasn’t been verified by an independent research firm. Treat it as a market pulse reading, not an audited benchmark.
What’s driving this? The competitive pressure from open-source models and Chinese AI providers has given Western frontier labs a commercial reason to price aggressively. That underlying dynamic is real even where the exact percentages aren’t yet confirmed.
For enterprise buyers, the practical question isn’t the percentage, it’s whether your current API contracts reflect 2026 pricing or 2024 pricing. That gap, whatever its precise size, is worth examining.