OpenAI's Token-Based API Pricing

OpenAI created a new pricing paradigm for AI: charging per token (roughly per word) processed, allowing developers to start at near-zero cost while scaling linearly with usage — and creating the infrastructure for the AI application economy.

Challenge

AI models are expensive to run, but traditional enterprise software pricing (per seat, per month) doesn't match how AI products are used. Charging too much upfront would kill developer experimentation. Charging too little would make the economics unsustainable. OpenAI needed a pricing model that worked for hobbyists and enterprises simultaneously.

Approach

OpenAI priced API access per 1,000 tokens (roughly 750 words), with different rates for input vs output and different model tiers (GPT-3.5 vs GPT-4). Developers could start with $5 in free credits, build and test applications, then scale costs linearly with actual usage. The usage-based model meant a startup with 100 users paid little while an enterprise with millions paid proportionally — no sales negotiation required. OpenAI also offered fine-tuning pricing (higher per-token for training) and introduced tiered rate limits that encouraged upgrades. The model allowed thousands of AI startups to build on OpenAI's infrastructure without upfront costs, creating an ecosystem that drove demand.

Results

• Revenue run-rate (2024): $2B+
• API users: Millions
• Applications built on OpenAI: Hundreds of thousands
• Valuation (2024): $80B+

Sources

• OpenAI pricing page and documentation
• Sam Altman interviews (various)
• OpenAI company announcements

The full record sits in the studio register.