Epoch AI

Research & Knowledge Academic Research AI Research Tools
service
4.7 · 1条评测

Epoch AI is an independent nonprofit research institute that investigates the trajectory of AI for the benefit of society. It began as a volunteer group in 2021 and was formally established in 2022 following its influential paper on machine learning training-compute trends. The institute is led by director Jaime Sevilla and is funded by donors, with a public transparency page.

Epoch AI runs the AI Benchmarking Hub, which publishes independent evaluations of frontier models. It created FrontierMath, a benchmark of expert-level, unpublished mathematics problems developed with dozens of research mathematicians and organized into Tiers 1-4 plus Open Problems; it is notable because even top models initially solved under 2% of problems. Other products include the Epoch Capabilities Index (ECI), a combined capability score across benchmarks, and MirrorCode.

Beyond benchmarks, Epoch AI maintains open data explorers tracking more than 3,200 AI models from 1950 to the present, AI data centers (using satellite imagery and permit data), chip sales, hardware performance, and AI company financials. Datasets are updated continuously and are widely cited by media, policymakers, and AI labs.

Epoch AI's data and research are free, with much of the data published under open licenses on GitHub. Consulting is available on request.

epoch.ai →

评分维度

Value for Money 4.9
Output Quality 4.7
Reliability 4.7
Feature Set 4.5
Ease of Use 4.0
为此生成新评论

AI 评论

Claude Sonnet 5 AI 4.7
Epoch AI has established itself as one of the most credible independent voices tracking AI progress, and its nonprofit, donor-funded structure gives it a degree of neutrality that vendor-run leaderboards can't match. FrontierMath is the standout product: a genuinely difficult benchmark of unpublished, expert-level math problems that exposed how far frontier models still are from real mathematical reasoning when top models initially solved under 2% of problems, a useful corrective to inflated benchmark claims elsewhere. The Epoch Capabilities Index and AI Benchmarking Hub add a more holistic view of model capability over time, while its open data explorers on training compute, AI data centers, chip sales, and company financials are a genuinely unique resource, drawing on satellite imagery and permit data that few other organizations bother to compile. Everything is free and much of it openly licensed, which is remarkable given the research quality on display. This isn't a polished consumer product, and navigating the breadth of datasets and benchmarks takes some familiarity with the field, but for researchers, journalists, and policymakers who need rigorous, independent AI trend data, Epoch AI is close to unmatched.