The AI Pricing Race to Zero

In early 2024, GPT-4 cost $30 per million input tokens. Today, models of comparable quality cost under $1. That's a 97% price drop in two years. Here's what happened and where pricing is heading.

The Price Collapse Timeline

The great AI price war started in earnest in mid-2024 when several things happened simultaneously:

• GPT-4o launched at $5/$15 — less than half the price of GPT-4 Turbo, with better performance.
• DeepSeek V2 showed that efficiency wins — by using a Mixture of Experts architecture, they achieved GPT-4 class performance at 1/10th the cost.
• Google made Gemini Flash nearly free — aggressively pricing their fastest model to gain market share.
• Open-source models reached GPT-4 level — Llama 3.1 70B matched GPT-4 on many benchmarks, available for free.

Where Are We Now?

The current pricing landscape looks remarkably different from even a year ago:

What This Means for You

If you're building with AI or choosing a model:

1. Don't overpay for tasks that cheap models handle well. Simple text generation, summarisation, and classification can often be done by $0.10/1M models.
2. Use premium models only where quality matters. Complex reasoning, creative writing, and coding tasks benefit from the best models.
3. Watch the open-source gap closing. Self-hosting is becoming cost-effective for high-volume use cases.
4. Expect prices to keep falling. Hardware improvements (Nvidia Blackwell, custom chips from Google/Amazon) will push costs lower.