AI model benchmark
Which AI model is most capable right now?
The Epoch Capabilities Index puts frontier AI models on one scale, built from dozens of benchmarks of different difficulty. Here is the current ranking, how fast the frontier is moving, and what the numbers mean when you are picking a model for real work.
Leaderboard
Current ECI ranking
Each bar is a model's ECI score. The thin line shows Epoch's 90% interval: when two models' lines overlap, the data cannot really tell them apart.
The frontier over time
How fast is the frontier moving?
Every dot is a model at its release date. The line steps up each time a new model sets the record. Hollow dots are open-weight models.
By lab
Best model from each lab
The highest-scoring model each lab has on the index today.
Open vs proprietary
How far behind are open-weight models?
Head to head
Compare two models
Pick any two models on the index. We show the score gap and whether it is bigger than the uncertainty.
Methodology
How the Epoch Capabilities Index works
Single benchmarks wear out. Once the best models score near 100%, a test stops telling them apart, and every model is tested on a different mix of benchmarks. The ECI solves both problems by putting models and benchmarks on one shared scale.
It uses Item Response Theory, the statistics behind standardized tests that compare students who sat different exams. Each benchmark gets a difficulty and a slope. Each model gets one capability number. A model scores well on a benchmark when its capability is high relative to that benchmark's difficulty. Epoch fits all of these values at once from every score it has, drawn from its own evaluation runs and from external leaderboards.
Collect scores
Results from dozens of benchmarks, from older ones like MMLU to hard current ones like FrontierMath.
Fit one scale
Hard benchmarks count for more. A model that does well on hard tests ranks higher than one that only aces easy ones.
Anchor it
The scale is pinned so Claude 3.5 Sonnet scores 130 and GPT-5 (medium) scores 150. The number itself has no units, only distance.
On November 4, 2026 Epoch moved the ECI to a Bayesian model, which gives more realistic uncertainty ranges. Frontier scores barely changed, but scores from before that date are not directly comparable, so this page always shows the current version. Full details are in Epoch's methodology notes and the paper A Rosetta Stone for AI Benchmarks.
What it means for your business
Reading the ranking before you pick a model
A few points is often a tie
If two models' intervals overlap, treat them as equal on general capability and decide on price, speed, privacy terms and how they handle your own tasks.
General is not specific
ECI blends many skills into one number. A model ranked fifth overall can be the best choice for coding, document extraction or customer support. Test it on your work.
The frontier moves every few months
Build so you can swap models without rebuilding the workflow. The model you pick today will not be the leader a year from now.
Open weights close the gap
Open-weight models trail the leaders by months, not years. For private data or high volume, running one yourself can be the right trade.
Free AI audit
Not sure which model fits your workflow?
We map your processes, test the right models on your own tasks, and tell you what to automate first.
FAQ
Common questions
What is the Epoch Capabilities Index?
The Epoch Capabilities Index (ECI) is a single score for general AI capability, published by the research group Epoch AI. It combines results from dozens of benchmarks using Item Response Theory, so models tested on different benchmarks can still be compared.
Which AI model has the highest ECI score?
What does an ECI score of 150 mean?
The scale is anchored so that Claude 3.5 Sonnet scores 130 and GPT-5 (medium) scores 150. A score has no units on its own. What matters is the distance between models and how that distance compares to the uncertainty range.
How often is this page updated?
We refresh the data from Epoch AI's published dataset and show the date of the latest refresh at the top of the page. New models usually appear within a few weeks of release, once enough benchmark results exist.
Why is my favorite model missing?
A model needs scores on enough benchmarks before Epoch can place it on the scale. Models that are only available to a few testers, or that were just released, may not be on the index yet.
Should I choose an AI model based on ECI alone?
No. ECI is a good guide to overall capability, but price, speed, data privacy terms, context length and performance on your own tasks matter as much. Use it to build a shortlist, then test the shortlist on real work.
Who runs the evaluations?
Epoch AI, an independent research organization. It runs some benchmarks itself and collects others from official leaderboards, and it publishes the data under a Creative Commons Attribution license. JMK Ventures does not run these evaluations; we visualize and explain them.
Sources and license
Data: Epoch AI, "Capabilities & benchmarking". Published online at epoch.ai. Retrieved from epoch.ai/benchmarks/use-this-data, . Licensed under CC BY 4.0. JMK Ventures selected, charted and summarized the data; Epoch AI does not endorse this page.
Methodology: Epoch AI, ECI methodology and Bayesian update (November 4, 2026).
Model names are trademarks of their owners. Scores describe performance under Epoch AI's evaluation conditions and may differ from vendor-reported results.