OpenAI cut prices 99%. Your SaaS raised them 30%. Who pays the difference?

In March 2023, a million GPT-4 tokens cost $30 USD. Today GPT-4o-mini costs $0.15 USD. AI inference is 99.5% cheaper. Meanwhile HubSpot, Salesforce and Microsoft Copilot raised prices. Let's do the math: who captures the savings?

AI inference quietly became a commodity

March 2023: OpenAI launched GPT-4 at $30 USD per million input tokens. November 2023: GPT-4 Turbo at $10. May 2024: GPT-4o at $5. July 2024: GPT-4o mini at $0.15. A 99.5% drop in 16 months. Anthropic, Google and DeepSeek did the same: Claude Haiku 3.5 at $0.80/M, Gemini Flash at $0.075/M, DeepSeek V3 at $0.14/M. What was a premium feature in 2023 is utility tap water today.

But your SaaS bill didn't drop. It went up.

While OpenAI was cutting prices 99%, HubSpot raised Marketing Hub Enterprise 8% in 2024 and another 5% in 2025 (Forbes Tech Council, August 2024). Salesforce repriced Einstein 1 Cloud +9% on Agentforce launch. Microsoft 365 Copilot holds firm at $30 USD/user/month since launch (no decrease despite underlying cost dropping an order of magnitude). The "generative AI" features these SaaS bundle are powered by OpenAI tokens that now cost two orders of magnitude less.

Who captures the savings?

When a SaaS company pays $0.15 USD to process what cost it $30 in 2023, the $29.85 savings go three places: 1) the SaaS's P&L, 2) new features they sell as 'AI-powered', 3) marketing. What does NOT drop: what they charge you. The asymmetry is legitimate — they're private businesses — but selling it as 'premium added value' is charging you for what's already a commodity. It's the exact Spotify model of charging for songs the artist gets pennies per play.

The new rule for 2026

If your company will pay for AI in 2026, pay direct to the maker (OpenAI, Anthropic, Google, DeepSeek) and integrate via custom software in your processes. Any SaaS selling you 'AI inside' as premium pricing justification is charging margin, not value. The math is simple: custom integration + direct APIs vs SaaS with bundled AI. 9 out of 10 times, custom wins by 70%+ and you keep the data, context and control. If you already have 50 users paying full suites just for the AI features, ROI of switching to custom pays back in under 8 months.