https://ai-news.ypetit.net/sources/epoch-ai/ Latest publications from Epoch AI, aggregated by AI News from the official source. en Tue, 10 Nov 2026 00:08:45 GMT https://epoch.ai/publications/inference-performance-indexhttps://epoch.ai/publications/inference-performance-indexMon, 09 Nov 2026 00:00:00 GMTresearchEpoch AI's Inference Performance Index, which goes beyond the underestimates of FLOP/s measures and scores chips on memory and communication https://epoch.ai/publications/huawei-horinger-china-compute-gaphttps://epoch.ai/publications/huawei-horinger-china-compute-gapFri, 06 Nov 2026 00:00:00 GMTresearchHuawei’s AI data center in Horinger, Inner Mongolia, uses Ascend chips and has an estimated AI computing capacity that is just 7% that of the world’s largest AI data center, SpaceX’s Colossus 2 in Tennessee. https://epoch.ai/data-insights/tpu-production-doublinghttps://epoch.ai/data-insights/tpu-production-doublingWed, 04 Nov 2026 00:00:00 GMTresearchBroadcom’s shipments of TPU by unit volume increased by about 65% quarter over quarter to roughly 800,000 chips, while compute output doubled, driven by a shift to more powerful chips. https://epoch.ai/publications/eci-bayesian-methodologyhttps://epoch.ai/publications/eci-bayesian-methodologyWed, 04 Nov 2026 00:00:00 GMTresearchA Bayesian model gives the ECI better-justified uncertainty estimates and lets us compare models directly. https://epoch.ai/data-insights/biorxiv-llms-vs-ai-toolshttps://epoch.ai/data-insights/biorxiv-llms-vs-ai-toolsFri, 30 Oct 2026 00:00:00 GMTresearch17% of biology preprints acknowledge LLM use, nearly matching specialized tools like AlphaFold. Epoch AI analyzes trends in AI acknowledgment in research. https://epoch.ai/publications/open-weight-models-patch-gaphttps://epoch.ai/publications/open-weight-models-patch-gapThu, 29 Oct 2026 21:40:00 GMTresearchGLM 5.3 turned published-but-unshipped Firefox security fixes into working one-click exploits for six releases, at roughly $1,000 each. https://epoch.ai/data-insights/ai-computing-capacity-additions-doublinghttps://epoch.ai/data-insights/ai-computing-capacity-additions-doublingThu, 29 Oct 2026 00:00:00 GMTresearchThe computing power of AI chips sold each quarter has grown about 2.5× per year since 2024, down from a surge in 2023. https://epoch.ai/latest/introducing-the-ai-chip-imports-explorerhttps://epoch.ai/latest/introducing-the-ai-chip-imports-explorerTue, 27 Oct 2026 00:00:00 GMTresearchOur explorer uses customs records to estimate national shares. Since 2022, 72% of AI compute has ended up in the US and about 10% in China, including chips China makes itself. https://epoch.ai/data-insights/open-closed-cyber-eci-gaphttps://epoch.ai/data-insights/open-closed-cyber-eci-gapFri, 16 Oct 2026 00:00:00 GMTresearchZ.ai’s GLM-5.3 is the top open-weight model on Epoch’s Cyber ECI, scoring where the closed-weight frontier stood 6–7 months earlier. https://epoch.ai/publications/ai-systems-are-improving-rapidly-at-slay-the-spire-and-its-sequelhttps://epoch.ai/publications/ai-systems-are-improving-rapidly-at-slay-the-spire-and-its-sequelWed, 14 Oct 2026 00:00:00 GMTresearchMonths ago, frontier models could barely beat Slay the Spire. Now they win 90%+ at the easiest setting and up to 20% at the hardest. Results from Epoch AI’s new SpireClimb benchmark. https://epoch.ai/publications/can-ai-automate-epochhttps://epoch.ai/publications/can-ai-automate-epochThu, 08 Oct 2026 00:00:00 GMTresearchFrontier models can handle well-defined tasks reliably, but fail in the more open-ended aspects that prevent full work automation. Explore Epoch AI's Automation Reports: realistic, open-ended tasks taken directly from our work, manually graded by our own standards. https://epoch.ai/data-insights/cyber-incidents-flat-since-fable-5https://epoch.ai/data-insights/cyber-incidents-flat-since-fable-5Wed, 07 Oct 2026 00:00:00 GMTresearchJune and September polls from Epoch AI/Ipsos show no detectable increase in the share of US adults experiencing cyber incidents https://epoch.ai/publications/ebr-bench-updatehttps://epoch.ai/publications/ebr-bench-updateWed, 07 Oct 2026 00:00:00 GMTresearchUpdates on EBR-bench, Epoch AI’s benchmark that tests models’ ability to learn from experience. GPT-6 Astra hit a perfect score on over half its EBR-bench attempts by exploiting one overpowered card, which we now ban. Plus multi-agent scaffold results. https://epoch.ai/publications/innovationevalhttps://epoch.ai/publications/innovationevalWed, 07 Oct 2026 00:00:00 GMTresearchEarly evidence from InnovationEval: No https://epoch.ai/publications/china-us-semiconductor-supply-chain-exposurehttps://epoch.ai/publications/china-us-semiconductor-supply-chain-exposureTue, 06 Oct 2026 00:00:00 GMTresearchChina is 2.7× as exposed to semiconductor supply shocks as the US, Epoch AI estimates. Epoch traces chips through input-output tables to model the impact of a Taiwan shock and decoupling. https://epoch.ai/publications/how-do-chinese-ai-companies-make-moneyhttps://epoch.ai/publications/how-do-chinese-ai-companies-make-moneyTue, 06 Oct 2026 00:00:00 GMTresearchChina's six leading AI firms earn about 10% of OpenAI and Anthropic's combined AI revenue. Epoch AI examines their five revenue streams: consumer apps, model access, enterprise and government solutions, licensing fees, and indirect monetization through cloud and advertising. https://epoch.ai/data-insights/openai-coding-agent-spendinghttps://epoch.ai/data-insights/openai-coding-agent-spendingMon, 05 Oct 2026 00:00:00 GMTresearchOpenAI researchers’ coding-agent usage, valued at API prices, has recently been doubling roughly every month, according to breakpoint fits. https://epoch.ai/publications/estimating-the-agent-populationhttps://epoch.ai/publications/estimating-the-agent-populationFri, 02 Oct 2026 00:00:00 GMTresearchMemory shipped through 2027 could run 33–171 million concurrent frontier-model agents, or billions using efficient open models. Epoch AI estimates inference capacity from HBM supply, serving benchmarks, and agent-hour costs. https://epoch.ai/latest/introducing-the-chatgpt-usage-explorerhttps://epoch.ai/latest/introducing-the-chatgpt-usage-explorerThu, 01 Oct 2026 00:00:00 GMTresearchMedian monthly messages per active user rose from 14 in 2023 to 36 in 2025, and the top 10% of users sent 63% of all prompts. Epoch AI's ChatGPT usage explorer draws on 8.3 million messages from 5,000 US YouGov panelists. https://epoch.ai/publications/furniture-assemblyhttps://epoch.ai/publications/furniture-assemblyWed, 23 Sep 2026 00:00:00 GMTresearchAI model scores on Epoch AI's IKEA furniture assembly benchmark jumped from 28% to 80% in 10 months, with open-weight models trailing closed-weight models by about 7 months. Epoch AI's test of visual and spatial reasoning across 60 photos from three furniture builds. https://epoch.ai/publications/the-plunging-price-of-thoughthttps://epoch.ai/publications/the-plunging-price-of-thoughtTue, 22 Sep 2026 00:00:00 GMTresearchEpoch AI measures how fast the cost of a given level of AI performance is falling across five benchmarks covering math, science and games of skill: about 47% per quarter, or 13x per year, since 2023, faster than electricity, compute, batteries or DNA sequencing ever fell. https://epoch.ai/data-insights/math-preprints-disclosed-ai-usehttps://epoch.ai/data-insights/math-preprints-disclosed-ai-useFri, 18 Sep 2026 00:00:00 GMTresearchAcknowledgments of AI use in arXiv math preprints rose from 4% in April 2026 to 25% in August, with 6% crediting AI with a substantial research contribution. https://epoch.ai/data-insights/malaysia-china-chip-smugglinghttps://epoch.ai/data-insights/malaysia-china-chip-smugglingThu, 17 Sep 2026 00:00:00 GMTresearchEpoch AI's analysis of trade data finds a pattern consistent with $3B worth of chips smuggled to China via Malaysia. China recorded $3.8B of server imports from Malaysia in 2024–25; Malaysia recorded only $0.6B of exports. https://epoch.ai/data-insights/astra-eci-breakdownhttps://epoch.ai/data-insights/astra-eci-breakdownWed, 16 Sep 2026 00:00:00 GMTresearchOpenAI's GPT-6 Astra tops the Epoch Capabilities Index (ECI) with a score of 166, ahead of Claude Fable 5.1 at 164 and GPT-5.6 Sol at 162. Its Math-ECI of 170 sets a new record, but on software engineering benchmarks its SWE-ECI of 164 still lags behind Fable 5.1's 167.