Every 2-4 months, the price of a given AI intelligence level is drop by at least 50%.

Analyzing the Economics of AI Progress: Dramatic Cost Reductions Every Few Months

In recent discussions across AI communities, a recurring concern has been the rising costs associated with AI inference. However, a closer examination of market data reveals a different narrative—one of rapid and substantial cost reductions over time. Let’s delve into the facts, backed by comprehensive benchmarking and price tracking, to understand this trend.

Understanding the Metrics: The Estimated Capability Index (ECI)

To quantify AI capabilities objectively, I’ve compiled data from the Epoch AI benchmarking suite, which integrates various state-of-the-art assessments—such as MMLU, GPQA, coding challenges, and reasoning tests—into a singular, composite score called the Estimated Capability Index (ECI). The ECI reliably interpolates between models, effectively capturing the nuanced differences in their overall intelligence levels.

Key Milestones in AI Development and Cost Trends

1. Launch of GPT-4 (March 14, 2023):
2. ECI score: 126 points

3. Cost to acquire this capability: approximately $37.50 (blended input/output cost)

4. Current Market (April 30, 2026):

5. Cost to reach an ECI of 126: roughly $0.13

This comparison illustrates a staggering 99.6% decrease in the cost of achieving the same level of AI capability in just over three years.

Similarly, earlier milestones further exemplify this trend:

• January 20, 2025: DeepSeek-R1 achieved an ECI of 140 with a minimum cost of $0.96.

• April 9, 2025: The release of Grok-3 mini reduced this cost to $0.26 for an ECI of 141—a 72% drop within three months.

• November 13, 2025: GPT-5.1 reached an ECI of 150 at a cost of $3.43.

• December 17, 2025: Google’s Gemini 3 Flash hit an ECI of 151, with costs falling further to $1.12, marking a 67% decrease over just over a month.

Visualizing Cost Declines Over Time

[Insert graphs indicating the exponential decline in inference costs for various ECI benchmarks over time]

This rapid progression suggests a clear pattern: in the current AI market, the expense to attain a given intelligence level halves every 2 to 4 months.

Implications and Future Outlook

Such explosive cost reductions challenge the conventional assumption that AI costs inevitably climb as capabilities improve. Instead, this data indicates the opposite: advances in model efficiency, optimization, and hardware acceleration are driving costs downward at an unprecedented rate.

If this trend persists, achieving what might be considered “human-level intelligence” could eventually approach negligible expenses. While the exact ECI that corresponds to human equivalence remains undefined, the trajectory implies that affordability and access to highly capable AI will become increasingly widespread and economical.

In summary, the market data strongly suggests that AI inference costs are decreasing exponentially, not increasing. This phenomenon opens exciting possibilities for democratizing AI applications and accelerating technological progress.

Sources for Data and Benchmarking:
– Epoch AI Trends
– Epoch AI ECI Graphs

By staying informed on these trends, stakeholders can better anticipate the future landscape of artificial intelligence and its potential impacts.

---

Note: Graphs illustrating the decline in inference costs over time are included for visual reference.