Epoch AI: Model Costs Are Falling Faster Than Any Past Technology
- Source
- x.com
- Date
Epoch AI's updated analysis measures what it costs to reach a fixed level of benchmark performance over time, rather than tracking list prices for individual models. Across 222 models and 11 benchmarks, the price of a given result has dropped roughly 47% each quarter since 2023. The decline is steepest for state-of-the-art results and for math tasks, which suggests the frontier itself is where efficiency gains land first. Epoch compares the rate to past transformative technologies and finds it faster than DNA sequencing, general compute, or batteries at their steepest. For teams budgeting agent workloads, the implication is that a capability too expensive today may cost a fraction of that within a year, so fixed long-term assumptions about model spend age quickly. This week's same-day price cuts from Anthropic and OpenAI are a single data point consistent with that trend.
AI is getting cheaper more quickly than any other transformative tech in history. At a given level of performance, cost has fallen ~47%/quarter since 2023. That’s 4× faster than DNA sequencing, 6× faster than compute, 18× faster than lithium batteries, and (up to 1973) 54× faster than electricity.
For one example of the plunging “cost of thought,” take OpenAI’s o3 model. Released in January 2025, it was the first LLM to score >25% on FrontierMath Tiers 1-3. The bill for achieving this mark: $0.55. This summer, GPT-5.6 Luna did the same for $0.0015. A 377-fold drop in 18 months.
Our estimates come from an expanded Epoch dataset on the performance of 222 AI models across up to 11 benchmarks. The dataset is built using a method borrowed from CAISI, which lets us estimate how the same model would perform under various budget constraints.
The speed of price drop varies by domain: slower on game-based puzzles, at 39–43% per quarter, faster on math problems, at 50–52% per quarter.
Context
Epoch AI reports that, at a fixed level of AI performance, cost has fallen about 47% per quarter since 2023. Epoch says that is 4 times faster than DNA sequencing, 6 times faster than compute and 18 times faster than lithium batteries, and calls AI cheaper more quickly than any other transformative technology in history.
Its example is OpenAI's o3, released in January 2025 as the first to score above 25% on FrontierMath Tiers 1-3, at a cost of $0.55 for that mark. Epoch says GPT-5.6 Luna did the same this summer for $0.0015, a 377-fold drop in 18 months.
The estimates come from an expanded Epoch dataset of 222 models on up to 11 benchmarks, using a method borrowed from CAISI to estimate how the same model would perform under different budget limits. The rate varies: 39 to 43% per quarter on game-based puzzles, 50 to 52% on math, and, averaged across five benchmarks, 66% per quarter for state-of-the-art performance, slowing to 32% two years later.
- LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
Checking sign-in…
Loading comments…