Top all-time · #296
Epoch AI
Company blog · epoch.ai/gradient-updatesaidata
Epoch AI is the company blog at epoch.ai, writing about ai and data. Since 2024, 174 different Hacker News readers have linked it 253 times, most often “How much energy does ChatGPT use?”. It is still writing: the last post was on 2026-10-30, and you can follow it by feed.
Rarely recommended · 4 of 40 sampled comments recommend it?
Why HN links it
Commenters cite Epoch AI's charts and analyses on AI benchmarks, compute and inference costs. One calls its falling-price-per-task analysis more defensible and grounded than the article it discusses.
Summarised by Claude from a sample of 40 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Epoch AI is a research institute investigating key trends and questions that will shape the trajectory and governance of Artificial Intelligence.
Most-linked posts?
- How much energy does ChatGPT use?12 people · 12 mentions · first linked 2025-04-29
- /data-insights/llm-inference-price-trends10 people · 17 mentions · first linked 2025-05-21
- AI is getting cheaper more quickly than any other transformative tech in history10 people · 11 mentions · first linked 2026-09-22
- AI Benchmarking Dashboard5 people · 5 mentions · first linked 2024-12-04
- Gemini tops leaderboard on research math problems5 people · 5 mentions · first linked 2024-12-22
- How has DeepSeek improved the Transformer architecture?4 people · 7 mentions · first linked 2025-01-18
- Global AI computing capacity is doubling every 7 months4 people · 4 mentions · first linked 2026-01-09
- Can AI companies become profitable?4 people · 4 mentions · first linked 2026-01-29
- /data-insights/openai-compute-spend3 people · 4 mentions · first linked 2025-10-14
- /trends3 people · 3 mentions · first linked 2025-04-22
What Hacker News says?
No i don't recommend those anymore. Currently I recommend Epoch's ECI. AA was kind of bad but recent changes are trending a little bit better. I think benchmarks are in a really bad state right now otherwise, I basically don't trust them, good related reading https://epoch.ai/data/benchmark-reviews-documentation/includ...
Comment linking /data/benchmark-reviews-documentation/included-benchmarks?utm_source=alphasignal
on “HackerTube: YouTube built from videos recommended by HN”Submitted How has DeepSeek improved the Transformer architecture? as a story · 258 points
HN discussionhttps://epoch.ai/data-insights/llm-apis-accuracy-runtime-tra... It is easier to get from 0% accurate to 99% accurate, than it is to get from 99% accurate to 99.9% accurate. This is like the classic 9s problem in SRE. Each nine is exponentially more difficult. How easy do we really think it will be for an LLM to get 100% accurate at physics, when we don't even know what 100% right is, and it's theoretically possible…
Comment linking /data-insights/llm-apis-accuracy-runtime-tradeoff
on “GPT-5”Submitted Most AI value will come from broad automation, not from R & D as a story · 205 points
HN discussionProbably more. 200 kcal (a shrinkflated bag of chips) is about 232 watt hours. A typical 4o query is 0.3 to 3 watt hours. https://epoch.ai/gradient-updates/how-much-energy-does-chatg...
Comment linking /gradient-updates/how-much-energy-does-chatgpt-use
on “GPT-5”Submitted Recent AI models struggled to match a human algorithmic innovation as a story · 93 points
HN discussionThat’s totally not true. https://epoch.ai/gradient-updates/how-has-deepseek-improved-...
Comment linking /gradient-updates/how-has-deepseek-improved-the-transformer-architecture
on “DeepSeek-R1”Submitted What's the largest software project AI can complete on its own? as a story · 19 points
HN discussion> How much VRAM and inference compute is required to run 3.1-70B vs 2-70B? We aren’t trying to mindlessly consume the same VRAM as last year and hope costs magically drop. We are noticing that we can get last year’s mid-level performance on this year’s low-end model, leading to cost savings at that perf level. The same thing happens next year, leading to a drop in cost at any given perf level over time. > For traini…
Comment linking /blog/trends-in-gpu-price-performance
on “OpenAI O3 breakthrough high score on ARC-AGI-PUB”Submitted Gate: AI and Automation Scenario Explorer as a story · 19 points
HN discussion> making the most interesting and challenging LLM benchmark so far. This[1] is currently the most challenging benchmark. I would like to see how O3 handles it, as O1 solved only 1%. 1. https://epoch.ai/frontiermath/the-benchmark
Comment linking /frontiermath/the-benchmark
on “OpenAI O3 breakthrough high score on ARC-AGI-PUB”Submitted Open-weight AI models can exploit the patch gap as a story · 13 points
HN discussion> Why do you think the world to date hasn't been taken over by evil genius mathematicians? A "mathematician" is a human who decided to spend their lives studying mathematics. Mathematicians also tend to be smart, but intelligence is innate, not acquired, so studying mathematics doesn't make you smarter. This makes it obvious why they don't rule the world - if you want to rule the world you'd want to focus on that (f…
Comment linking /eci
on “Sharing AI progress in mathematics”Submitted What went into training DeepSeek-R1? as a story · 13 points
HN discussion> what about the fact frontier labs are spending more compute on viral AI video slop and soon-to-be-obsoleted workplace usecases than research? That's a bold claim, please cite your sources. It's hard to find super precise sources on this for 2025, but epochAI has a pretty good summary for 2024. (with core estimates drawn from the Information and NYT https://epoch.ai/data-insights/openai-compute-spend The most relev…
Comment linking /data-insights/openai-compute-spend
on “Andrej Karpathy – It will take a decade to work through the issues with agents”Submitted Chinese AI models have lagged the US frontier by 7 months on average since 2023 as a story · 9 points
HN discussionCurious you left out Frontier Math’s statement that they provided 300 questions plus answers, and another holdback set of 50 questions without answers, to allay this concern. [0] We can assume they’re lying too but at some point “everyone’s bad because they’re lying, which we know because they’re bad” gets a little tired. 0. https://epoch.ai/blog/openai-and-frontiermath
Comment linking /blog/openai-and-frontiermath
on “GPT-4.5”Submitted Trends in AI Supercomputers as a story · 7 points
HN discussionhttps://epoch.ai/benchmarks/simplebench Just do it. I use a service where I have access to all SOTA models and many open sourced models, so I change models within chats, using MCPs eg start a chat with opus making a search with perplexity and grok deepsearch MCPs and google search, next query is with gpt 5 thinking Xhigh, next one with gemini 3 pro, all in the same conversation. It's fantastic! I can't imagine what…
Comment linking /benchmarks/simplebench
on “Gemini 3 Flash: Frontier intelligence built for speed”Submitted AI is getting cheaper more quickly than any other transformative tech in history as a story · 6 points
HN discussionThis one is more powerful than openai models, including gpt 5.2 (which is worse on various benchmarks than 5.1 which is worse than 5.1, and that's where 5.2 was using XHIGH, whiulst the others were on high eg: https://youtu.be/4p73Uu_jZ10?si=x1gZopegCacznUDA&t=582 ) https://epoch.ai/benchmarks/simplebench
Comment linking /benchmarks/simplebench
on “Gemini 3 Flash: Frontier intelligence built for speed”Submitted The Plunging Price of Thought as a story · 5 points
HN discussionhttps://epoch.ai/eci?view=graph&tab=release-date is at least an alternative. They are slower to update, but entirely anecdotally I find the numbers are aligned with my personal observations a bit better than AA. But I am also highly aware this is some pretty confirmation-bias heavy ares of meta-benchmark picking...
Comment linking /eci?view=graph&tab=release-date
on “Kimi K3.1 Pro and K3.1 Pro Highspeed”Submitted Can AI companies become profitable? as a story · 5 points
HN discussionAccording to Epoch AI: > The cost of achieving a given level of AI performance has fallen about 47% per quarter since 2023, or 13× per year. [0] 0. https://epoch.ai/publications/the-plunging-price-of-thought
Comment linking /publications/the-plunging-price-of-thought
on “Claude Haiku 5.5”Submitted AGI could drive wages below subsistence level as a story · 5 points
HN discussion- nl · 2026-09-02
> we know deprecation cycle of DC hardware leans towards tulips i.e. <10 years (very generous) vs 20+ years for fiber layout [1] is a reasonable discussion of DC cost models, which calculates depreciation as part of the annual cost. > here is the rest of the order magnitude more $$$ that justifies existing spent relative to time frame coming from? That money comes from long term debt (ie bonds by public companies[3]…
Comment linking /data-insights/ai-datacenter-cost-breakdown
on “How accurate have Ed Zitron's AI skeptic predictions been?” Submitted Can AI Automate Epoch? as a story · 4 points
HN discussionThe data covers just ~10–20% of global aggregate AI supercomputer performance as of March 2025. And then there's this: https://epoch.ai/data/gpu-clusters-documentation/coverage > The coverage of Chinese companies is particularly poor. Our average coverage of 8 major companies is 15%. The bigger issue is that if you're interested in frontier training, the national aggregate doesn't tell you much. The question is whet…
Comment linking /data/gpu-clusters-documentation/coverage
on “Everyone should slow down AI development except for me”Submitted How many AI agents could run on the AI chips shipped through 2027? as a story · 4 points
HN discussion
