{"site":{"name":"ComputePulse","tagline":"AI and compute, live: model releases, research volume, benchmarks, GPU prices, token throughput, and the compute buildout.","domain":"computepulse.pages.dev","userAgent":"ComputePulse/0.1 (+https://computepulse.bristleconeanalytics.com; vital-signs dashboard)"},"instruments":[{"kind":"metric","order":"desc","staleAfter":"12h","seriesMode":"merge","unit":"models/24h","decimals":0,"history":720,"needsKey":null,"higherIsWorse":false,"describe":"Model repositories created on the Hugging Face Hub in the last 24 hours — the broadest available pulse of how fast open model publishing is moving.","id":"hf-new-models-24h","section":"Vitals","title":"New models published on Hugging Face","cadence":"6h","source":{"name":"Hugging Face Hub API","url":"https://huggingface.co/api/models?sort=createdAt&direction=-1","license":"Hub metadata, free/keyless API"},"thresholds":null},{"kind":"metric","order":"desc","staleAfter":"12h","seriesMode":"merge","unit":"of 3","decimals":0,"history":288,"needsKey":null,"higherIsWorse":true,"describe":"How many of OpenAI, Anthropic and Hugging Face are currently self-reporting anything other than fully operational.","id":"ai-providers-degraded","section":"Vitals","title":"AI providers reporting degradation","cadence":"6h","source":{"name":"OpenAI / Anthropic / Hugging Face status pages","url":"https://status.openai.com/","license":"public status endpoints"},"thresholds":[{"level":"serious","gte":2},{"level":"warning","gte":1}]},{"kind":"metric","order":"desc","staleAfter":null,"seriesMode":"merge","unit":"models","decimals":0,"history":240,"needsKey":null,"higherIsWorse":false,"describe":"How many distinct LLMs are commercially servable through a single aggregator right now — a proxy for how crowded the inference market has become.","id":"openrouter-models","section":"Models","title":"Models routable on OpenRouter","cadence":"6h","source":{"name":"OpenRouter","url":"https://openrouter.ai/api/v1/models","license":"public API, keyless"},"thresholds":null},{"kind":"events","order":"desc","staleAfter":"12h","seriesMode":"merge","unit":"","decimals":1,"history":120,"needsKey":null,"higherIsWorse":true,"describe":"The newest public model repos from the labs that set the frontier (Meta, Google, Qwen, DeepSeek, Mistral, NVIDIA, AI2, xAI, Moonshot, Z.ai and friends).","id":"hf-lab-releases","section":"Models","title":"Latest model releases from major labs","cadence":"6h","source":{"name":"Hugging Face Hub API","url":"https://huggingface.co/models?sort=created","license":"Hub metadata, free/keyless API"},"thresholds":null},{"kind":"metric","order":"desc","staleAfter":"5d","seriesMode":"merge","unit":"papers/day","decimals":0,"history":240,"needsKey":null,"higherIsWorse":false,"describe":"Daily rate of new cs.AI / cs.LG / cs.CL submissions, averaged over a complete 7-day window ending three days ago (arXiv indexes submissions with that much lag).","id":"arxiv-ai-papers","section":"Research","title":"AI preprints posted to arXiv","cadence":"6h","source":{"name":"arXiv API","url":"https://info.arxiv.org/help/api/index.html","license":"arXiv API Terms of Use — metadata free to reuse with attribution"},"thresholds":null},{"kind":"events","order":"desc","staleAfter":null,"seriesMode":"merge","unit":"","decimals":1,"history":120,"needsKey":null,"higherIsWorse":true,"describe":"The arXiv preprints the ML community is actually upvoting today — a curated counterweight to raw submission volume.","id":"hf-daily-papers","section":"Research","title":"Papers of the day","cadence":"6h","source":{"name":"Hugging Face Daily Papers","url":"https://huggingface.co/papers","license":"Hub metadata, free/keyless API"},"thresholds":null},{"kind":"metric","order":"desc","staleAfter":"120d","seriesMode":"merge","unit":"models","decimals":0,"history":400,"needsKey":null,"higherIsWorse":false,"describe":"Cumulative count of publicly known models trained with at least 10²⁵ floating-point operations — the compute level at which the EU AI Act and US reporting rules consider a training run frontier-scale.","id":"epoch-frontier-models","section":"Frontier & benchmarks","title":"Models trained above 10²⁵ FLOP","cadence":"1d","source":{"name":"Epoch AI — Notable AI Models","url":"https://epoch.ai/data/notable-ai-models","license":"CC BY 4.0"},"thresholds":null},{"kind":"metric","order":"desc","staleAfter":"720d","seriesMode":"merge","unit":"H100-equivalents","decimals":0,"history":200,"needsKey":null,"higherIsWorse":false,"describe":"Size of the biggest AI training cluster Epoch has confirmed is operational, normalised to H100-equivalent chips. The series shows only record-breaking clusters, so it is the historical high-water mark of hardware concentration.","id":"epoch-largest-cluster","section":"Frontier & benchmarks","title":"Largest known AI cluster","cadence":"1d","source":{"name":"Epoch AI — Data on AI Supercomputers","url":"https://epoch.ai/data/ai-supercomputers","license":"CC BY 4.0"},"thresholds":null},{"kind":"metric","order":"desc","staleAfter":null,"seriesMode":"merge","unit":"index","decimals":1,"history":240,"needsKey":"ARTIFICIAL_ANALYSIS_API_KEY","higherIsWorse":false,"describe":"Best score any model currently posts on the Artificial Analysis Intelligence Index, a composite of reasoning, coding and knowledge benchmarks — the single number that tracks the capability frontier.","id":"aa-intelligence-index","section":"Frontier & benchmarks","title":"Top Intelligence Index score","cadence":"6h","source":{"name":"Artificial Analysis","url":"https://artificialanalysis.ai/","license":"free API key required; see their terms for redistribution limits"},"thresholds":null},{"kind":"metric","order":"desc","staleAfter":"12h","seriesMode":"merge","unit":"$/GPU-hr","decimals":3,"history":720,"needsKey":null,"higherIsWorse":true,"describe":"Lowest advertised on-demand price for one H100 SXM on Vast.ai — a floor set by whichever independent host is cheapest right now. Vast.ai is a decentralised marketplace of smaller and hobbyist hosts, so a single cheap listing can pull this down even when vetted enterprise capacity (CoreWeave, Lambda, the hyperscalers) is tight. Read it as directional pressure on the loose end of the GPU market, not a market-wide clearing price.","id":"gpu-h100-spot","section":"Usage & prices","title":"Cheapest rentable H100","cadence":"6h","source":{"name":"Vast.ai marketplace","url":"https://cloud.vast.ai/create/","license":"public console API, keyless"},"thresholds":null},{"kind":"metric","order":"desc","staleAfter":null,"seriesMode":"merge","unit":"$/M tokens","decimals":3,"history":240,"needsKey":null,"higherIsWorse":true,"describe":"Median list price to send a million input tokens to a paid model with at least a 128k context window — the going rate for serious LLM work.","id":"openrouter-price-longctx","section":"Usage & prices","title":"Median input price, long-context models","cadence":"6h","source":{"name":"OpenRouter","url":"https://openrouter.ai/models","license":"public API, keyless"},"thresholds":null},{"kind":"metric","order":"desc","staleAfter":"7d","seriesMode":"merge","unit":"downloads/week","decimals":0,"history":400,"needsKey":null,"higherIsWorse":false,"describe":"Combined weekly npm installs of the five SDKs most JavaScript apps use to call a model (OpenAI, Anthropic, Google GenAI, LangChain, Vercel AI) — a demand signal from the builder side rather than the lab side.","id":"npm-ai-sdk-downloads","section":"Ecosystem","title":"AI SDK downloads from npm","cadence":"1d","source":{"name":"npm registry download API","url":"https://github.com/npm/registry/blob/master/docs/download-counts.md","license":"public API, keyless"},"thresholds":null},{"kind":"events","order":"desc","staleAfter":"12h","seriesMode":"merge","unit":"","decimals":1,"history":120,"needsKey":null,"higherIsWorse":true,"describe":"Every incident the big model providers have opened on their own status pages, newest first, with their own severity rating.","id":"ai-incidents","section":"Incidents","title":"AI platform incidents","cadence":"6h","source":{"name":"OpenAI / Anthropic / Google Cloud status pages","url":"https://status.anthropic.com/","license":"public status endpoints"},"thresholds":null},{"kind":"events","order":"desc","staleAfter":"12h","seriesMode":"merge","unit":"","decimals":1,"history":150,"needsKey":null,"higherIsWorse":true,"describe":"What the labs themselves said this week — model launches, safety posts and research write-ups, straight from their own feeds.","id":"lab-announcements","section":"Announcements","title":"Lab announcements","cadence":"6h","source":{"name":"OpenAI / DeepMind / Google / Hugging Face / Mistral blogs","url":"https://openai.com/news/","license":"publisher RSS feeds"},"thresholds":null}]}