# AI performance costs drop faster than any previous tech

Digest AI · Marketing & Small Business · published 2026-09-24T16:00:18Z

Canonical: https://digestai.news/story/ai-performance-costs-drop-faster-than-any-previous-tech

## Summary

Epoch AI, a research organization tracking AI trends, reports that the cost of reaching a fixed level of performance on select benchmarks has dropped by about 47% per quarter since 2023. This is an average rate of decline, and Epoch suggests this figure does not reflect progress in pure algorithmic efficiency or real-world productivity costs. MIT researchers using pricing data from Artificial Analysis have found similar trends but with a slightly lower annual reduction in cost attributed to algorithmic gains alone (about 3x per year). OpenAI's recent launches of GPT-6 Sol and Luna models, which are even cheaper than previous iterations, have further widened the gap between older and newer AI performance benchmarks. Epoch bases its analysis on five benchmarks covering math, science, and logic puzzles, noting that while some gains in efficiency can be attributed to better algorithms, other improvements come from using more compute per task. The researchers suggest that a single score may blend multiple factors such as training quality, data availability, architecture, and test-time computing power. Epoch's findings indicate that not all performance gains are efficiency gains; some models improve by simply running harder problems with more computational resources. Additionally, the research highlights that benchmark progress alone does not always equate to better real-world utility, as AI companies might optimize for well-known tests rather than practical applications. Therefore, Epoch recommends considering multiple factors beyond simple cost-per-token comparisons when choosing an AI model for a specific task.

## Key points

- AI performance costs have dropped by about 47% per quarter since 2023
- Epoch AI's analysis suggests this does not reflect pure algorithmic efficiency gains
- MIT researchers found similar trends but with slightly lower annual reduction in cost attributed to algorithmic gains (about 3x per year)

## Why it matters

This trend of falling AI performance costs is significant for the industry as it could lead to more widespread adoption and use of advanced AI technologies, potentially lowering barriers to entry for businesses and researchers.

## Sources

1. [AI performance costs are falling faster than those of any previous technology](https://the-decoder.com/ai-performance-costs-are-falling-faster-than-those-of-any-previous-technology) (The Decoder, 2026-09-24)

Part of the developing story: [AI Wage Gap and Performance Surge](https://digestai.news/thread/study-links-ai-exposure-to-6-7-slower-real-wage-growth-for-us-workers) (2 stories)

## Cite

Digest AI, "AI performance costs drop faster than any previous tech", 24 September 2026, https://digestai.news/story/ai-performance-costs-drop-faster-than-any-previous-tech

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/ai-performance-costs-drop-faster-than-any-previous-tech.json