{"id":315,"date":"2026-05-08T23:18:55","date_gmt":"2026-05-08T23:18:55","guid":{"rendered":"https:\/\/jua.ai\/articles\/ai-weather-model-benchmarks-2026\/"},"modified":"2026-07-11T05:00:35","modified_gmt":"2026-07-11T05:00:35","slug":"ai-weather-model-benchmarks-2026","status":"publish","type":"post","link":"https:\/\/jua.ai\/articles\/ai-weather-model-benchmarks-2026\/","title":{"rendered":"EPT-2 Leads AI Weather Model Benchmarks on Energy Variables"},"content":{"rendered":"<p><em>Written by: Olivier Lam, Physical AI Team, Jua.ai AG | Last updated: July 10, 2026<\/em><\/p>\n<h2 id=\"key-takeaways\">Why EPT-2 Matters for Energy Trading in 2025<\/h2>\n<ul>\n<li>EPT-2 leads all tested AI and traditional models on the four energy-critical variables (10 m wind, 100 m wind, 2 m temperature, SSRD) across the full 0\u2013240 h range in the 2025 StationBench evaluation against more than 10,000 real ground stations.<\/li>\n<li>EPT-2e, the ensemble variant, outperforms the 50-member ECMWF ENS mean on both RMSE and CRPS at virtually every lead time, and delivers superior probabilistic skill with only 10 members.<\/li>\n<li>Inference costs for EPT-2 are roughly four orders of magnitude lower than traditional NWP (~0.25 kWh vs. 8,400 kWh), which enables up to 24 updates per day and forecasts that finish 2.5 hours ahead of competing runs.<\/li>\n<li>EPT-2\u2019s physics-constrained architecture avoids the systematic biases seen in other AI models on extreme events and maintains accuracy on wind ramps, temperature extremes, and solar variability that directly affect energy P&amp;L.<\/li>\n<li>Book a demo with Jua to run EPT-2 head-to-head against your current provider on your own region and variables.<\/li>\n<\/ul>\n<h2>How StationBench Tests Real-World Forecast Quality<\/h2>\n<p>StationBench is an open-source evaluation framework that scores forecast models against raw observations from more than 10,000 surface stations globally. No post-processing is applied and no station fine-tuning is permitted. The result is a like-for-like comparison on the variables and lead times that matter operationally, not on gridded reanalysis fields that can be gamed by spatial smoothing.<\/p>\n<p>The 2025 evaluation covers deterministic skill via RMSE (root mean square error, the average magnitude of forecast error in physical units) and probabilistic skill via CRPS (continuous ranked probability score, a proper scoring rule that rewards both accuracy and calibration in ensemble forecasts). Lead times run from 0 to 240 hours. The model set includes EPT-2, EPT-2e, ECMWF HRES, ECMWF ENS, Microsoft Aurora, and GFS GraphCast, among others.<\/p>\n<p>Headline findings from <a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>:<\/p>\n<ul>\n<li>EPT-2 outperforms ECMWF HRES on every lead time across 10 m wind, 100 m wind, 2 m temperature, and SSRD.<\/li>\n<li>EPT-2e beats the 50-member ECMWF ENS mean on both RMSE and CRPS at virtually every lead time.<\/li>\n<li>EPT-2 beats Microsoft Aurora on 10 m wind, 100 m wind, and 2 m temperature across the full 0\u2013240 h range. Aurora has no SSRD output, so EPT-2 wins that variable by default.<\/li>\n<\/ul>\n<h2>EPT-2 Benchmark 2025: Core Performance on Energy Variables<\/h2>\n<p>EPT-2 is the deterministic flagship of the EPT family. It is global, has a 20-day horizon, runs four times per day, and natively forecasts at any lead time without rolling forward in fixed increments. Aurora and most peer models are trained on a fixed 6-hour grid and roll forward in 6-hour steps, which compounds error at longer lead times. EPT-2 avoids this roll-forward error.<\/p>\n<p>Performance on energy-critical variables, per <a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>:<\/p>\n<ul>\n<li><strong>10 m wind speed:<\/strong> EPT-2 leads ECMWF HRES and Microsoft Aurora across the full 0\u2013240 h range on RMSE.<\/li>\n<li><strong>100 m wind speed (hub height):<\/strong> EPT-2 leads ECMWF HRES and Microsoft Aurora across the full 0\u2013240 h range on RMSE. Aurora produces 100 m wind output.<\/li>\n<li><strong>2 m temperature:<\/strong> EPT-2 leads ECMWF HRES and Microsoft Aurora across the full 0\u2013240 h range. Aurora closes the gap beyond approximately 130 h but does not surpass EPT-2.<\/li>\n<li><strong>Surface solar radiation (SSRD):<\/strong> EPT-2 leads ECMWF HRES across the full 0\u2013240 h range. Aurora publishes no SSRD output.<\/li>\n<\/ul>\n<p>EPT-2 was trained on 8 \u00d7 H100 GPUs over 10 days, on more than 5 petabytes of weather and climate data from over 120 distinct sources, including geostationary and polar-orbiting satellites, SYNOP and METAR surface station networks, national radar networks, ocean buoys, ERA5 reanalysis, and operational ECMWF HRES initial-condition fields. This comprehensive training foundation enables EPT-2 to compete directly with the industry\u2019s gold standard.<\/p>\n<h2>AI vs ECMWF HRES 2025: Shifting the Benchmark Hierarchy<\/h2>\n<p>StationBench results show an AI model leading HRES on all four primary energy variables simultaneously across the full 0\u2013240 h lead-time range. This marks a structural shift in how the industry evaluates operational forecast quality.<\/p>\n<p>The comparison is conducted on raw station observations with no post-processing, which creates the most conservative possible test for an AI model. HRES benefits from decades of physical parameterisation tuning, while EPT-2 learns the governing dynamics directly from observational data. The result, per <a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>, is that EPT-2 outperforms HRES at every lead time on all four energy-critical variables.<\/p>\n<p>Jua for Energy does not replace ECMWF. Serious customers keep their ECMWF subscription and run Jua for Energy alongside it. ECMWF AIFS, ECMWF\u2019s own AI model, runs on the Jua platform as a guest model. The 2025 StationBench results change the hierarchy and turn the comparison into a genuine choice rather than a foregone conclusion.<\/p>\n<h2>GraphCast vs Aurora vs EPT-2: Side-by-Side Energy Metrics<\/h2>\n<p>The three most-cited AI weather models in operational energy contexts are Google DeepMind\u2019s GFS GraphCast, Microsoft Aurora, and Jua\u2019s EPT-2. The table below summarises head-to-head performance on energy-critical variables based on the 2025 StationBench evaluation (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>) and the EPT-1.5 report (<a href=\"https:\/\/arxiv.org\/abs\/2410.15076\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2410.15076<\/a>).<\/p>\n<table>\n<thead>\n<tr>\n<th>Variable (0\u2013240 h RMSE)<\/th>\n<th>EPT-2<\/th>\n<th>Microsoft Aurora<\/th>\n<th>GFS GraphCast<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>10 m wind speed<\/td>\n<td>Leads all models across full range (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>)<\/td>\n<td>Second, loses to EPT-2 across full 0\u2013240 h range (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>)<\/td>\n<td>EPT-1.5 outperforms GraphCast on European wind (<a href=\"https:\/\/arxiv.org\/abs\/2410.15076\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2410.15076<\/a>)<\/td>\n<\/tr>\n<tr>\n<td>100 m wind speed (hub height)<\/td>\n<td>Leads all models across full range (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>)<\/td>\n<td>Aurora produces 100 m wind output (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>)<\/td>\n<td>Not evaluated at 100 m in StationBench 2025<\/td>\n<\/tr>\n<tr>\n<td>2 m temperature<\/td>\n<td>Leads all models across full range, Aurora closes gap beyond ~130 h but does not surpass (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>)<\/td>\n<td>Competitive beyond ~130 h, loses to EPT-2 overall (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>)<\/td>\n<td>EPT-1.5 outperforms GraphCast on European temperature (<a href=\"https:\/\/arxiv.org\/abs\/2410.15076\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2410.15076<\/a>)<\/td>\n<\/tr>\n<tr>\n<td>Surface solar radiation (SSRD)<\/td>\n<td>Leads ECMWF HRES across full range, only model with SSRD output (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>)<\/td>\n<td>No SSRD output published (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>)<\/td>\n<td>No SSRD output in StationBench 2025 evaluation<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>Beyond accuracy, the product-level differences affect day-to-day operations. EPT-2 produces forecasts at native any-\u0394t, which means arbitrary lead times without rolling forward in fixed increments. Aurora and GraphCast are trained on fixed 6-hour grids and roll forward in 6-hour steps, which compounds error. EPT-2 inference runs approximately 25% faster than Aurora, per <a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>. Neither Aurora nor GraphCast ships a productised ensemble equivalent to EPT-2e.<\/p>\n<p><a href=\"https:\/\/meetings-eu1.hubspot.com\/guett\/energy-trading?uuid=d780665f-ff71-439c-addf-c80e49af0627\" target=\"_blank\"><strong>Compare EPT-2, Aurora, and GraphCast side-by-side<\/strong><\/a> on your region and variables on the Jua platform.<\/p>\n<h2>Computational Efficiency of AI Weather Models<\/h2>\n<p>The economics of forecast frequency are determined by inference cost. A single traditional NWP simulation consumes approximately 8,400 kWh and costs \u20ac1,000\u2013\u20ac20,000 to run on HPC infrastructure, taking one to two hours per cycle. That cost ceiling caps ECMWF HRES at two to four global runs per day and has constrained the energy industry for forty years.<\/p>\n<p>A single EPT-2 inference runs on a single GPU in minutes, at approximately 0.25 kWh and $0.20\u2013$15 per simulation, per <a href=\"https:\/\/nebius.com\/customer-stories\/jua\" target=\"_blank\">Jua\u2019s infrastructure case study<\/a>. This represents a cost delta of roughly four orders of magnitude compared to traditional NWP. The efficiency advantage extends to training as well. EPT-2 was trained on 8 \u00d7 H100 GPUs over 10 days, while Microsoft Aurora required 32 \u00d7 A100 GPUs over 18 days, which means four times fewer GPUs and a substantially shorter training cycle for EPT-2, per <a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>.<\/p>\n<table>\n<thead>\n<tr>\n<th>Operational Metric<\/th>\n<th>EPT-2 \/ EPT-2e (Jua)<\/th>\n<th>ECMWF HRES \/ ENS<\/th>\n<th>Aurora \/ GraphCast<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Inference energy per run<\/td>\n<td>~0.25 kWh on single GPU (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>)<\/td>\n<td>~8,400 kWh on HPC (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>)<\/td>\n<td>Similar order of magnitude to EPT-2 for inference, EPT-2 ~25% faster than Aurora (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>)<\/td>\n<\/tr>\n<tr>\n<td>Inference cost per run<\/td>\n<td>~$0.20\u2013$15 per simulation (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>)<\/td>\n<td>\u20ac1,000\u2013\u20ac20,000 per simulation (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>)<\/td>\n<td>Research-grade, no published operational cost<\/td>\n<\/tr>\n<tr>\n<td>Update frequency (operational)<\/td>\n<td>Up to 24\u00d7\/day (EPT-2 RR), EPT-2e 4\u00d7\/day (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>)<\/td>\n<td>2\u20134\u00d7\/day (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>)<\/td>\n<td>Typically 4\u00d7\/day, no productised operational schedule<\/td>\n<\/tr>\n<tr>\n<td>Spatial resolution (native forecast)<\/td>\n<td>Up to 5 km (EPT-2 HRRR, Europe) (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>)<\/td>\n<td>9 km (HRES), ~9 km (ENS since the June 2023 IFS Cycle 48r1 upgrade)<\/td>\n<td>~25 km at published resolution (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>)<\/td>\n<\/tr>\n<tr>\n<td>Productised ensemble<\/td>\n<td>EPT-2e, beats 50-member ECMWF ENS mean on RMSE and CRPS at virtually every lead time (<a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>)<\/td>\n<td>ENS, 50-member gold standard for probabilistic NWP<\/td>\n<td>No productised ensemble equivalent<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>The practical consequence for energy trading is direct. Between the two to four daily NWP runs, traders are looking at stale numbers. EPT-2 RR updates up to 24 times per day. EPT-2 HRRR delivers the same high-cadence output at up to 5 km resolution over Europe. A typical Jua run also completes approximately 2.5 hours ahead of competing operational runs at the same cycle, so customers see the next forecast before the market does.<\/p>\n<h2>AI Weather Models on Extreme Events<\/h2>\n<p>Extreme weather events such as wind ramps, cold snaps, heat waves, and heavy precipitation are the highest-stakes forecast moments for energy traders. They are also the scenarios where AI weather models have historically shown the most variance in performance.<\/p>\n<p>Independent research published in <a href=\"https:\/\/agupubs.onlinelibrary.wiley.com\/doi\/10.1029\/2025GL119740\" target=\"_blank\" rel=\"noindex nofollow\">Geophysical Research Letters (2025)<\/a> found that FourCastNet V2 Small and Pangu-Weather produce cold-biased boreal winter land temperature forecasts relative to ERA5, with global mean biases of \u22120.35 K and \u22120.26 K at 2-day lead times and \u22120.45 K and \u22120.07 K at 9-day lead times. The same study found that the hottest 10% of 9-day FourCastNet forecasts are on average 0.91 K colder than ERA5, a systematic underestimation of heat extremes that would directly affect gas demand and power price forecasts. Separate work by Z. Zhang et al. (2025) and Kent et al. (2025) found that AI weather and climate models perform worse than traditional numerical models when predicting record-breaking temperature extremes.<\/p>\n<p>EPT-2\u2019s physics-constrained architecture addresses this class of failure directly. EPT is a spatiotemporal transformer foundation model trained on observational physics. Its outputs respect the conservation laws of mass, momentum, and energy that govern the real atmosphere. The architecture cannot produce outputs that violate those laws in the way a generic transformer applied naively to physics would. The StationBench evaluation against more than 10,000 real ground stations, with no post-processing, is the most conservative available test of this claim, and EPT-2 leads on all four energy-critical variables, per <a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a>.<\/p>\n<h2>Frequently Asked Questions<\/h2>\n<h3>How StationBench Differs from WeatherBench 2<\/h3>\n<p>WeatherBench 2 evaluates forecast models against gridded reanalysis fields, primarily ERA5, interpolated to a regular latitude-longitude grid. This approach is computationally convenient but introduces a systematic bias. Models that have been trained on ERA5 or that smooth outputs spatially can score well on WeatherBench 2 without performing well on real observations. StationBench evaluates forecast models directly against raw observations from more than 10,000 real ground stations worldwide, with no post-processing or station fine-tuning applied. The result is a like-for-like test of operational forecast quality at the locations where energy assets actually sit, such as wind turbines, solar farms, and load centers, rather than on a smoothed global grid. For energy trading applications, StationBench is the more operationally relevant benchmark.<\/p>\n<h3>Current Leader on 100 m Wind and Solar Radiation<\/h3>\n<p>EPT-2 leads on both variables in the 2025 StationBench evaluation. On 100 m wind speed, the hub height relevant to most modern onshore and offshore wind turbines, EPT-2 outperforms ECMWF HRES at every lead time. Microsoft Aurora produces 100 m wind output. On surface solar radiation (SSRD), EPT-2 outperforms ECMWF HRES across the full 0\u2013240 h range and is the only major AI weather model to publish SSRD output at all, as Aurora has no SSRD output. For energy traders and asset operators whose P&amp;L depends on wind generation and solar dispatch, EPT-2 is the only AI model that covers both variables with documented benchmark leadership.<\/p>\n<h3>EPT-2 Inference Cost Advantage over Traditional NWP<\/h3>\n<p>A single EPT-2 inference runs on a single GPU in minutes, at approximately 0.25 kWh and $0.20\u2013$15 per simulation. A single traditional NWP simulation, such as an ECMWF HRES run, consumes approximately 8,400 kWh and costs \u20ac1,000\u2013\u20ac20,000 on HPC infrastructure, taking one to two hours per cycle. The cost delta is roughly four orders of magnitude. This asymmetry makes EPT-2 RR\u2019s 24-runs-per-day cadence economically viable, while the same update frequency would cost tens of millions of dollars per day using traditional NWP infrastructure. For energy traders, the operational consequence is that stale numbers between NWP runs are no longer a structural constraint, because EPT-2 RR refreshes up to 24 times per day and EPT-2 HRRR delivers high-cadence output at up to 5 km resolution over Europe.<\/p>\n<h3>How EPT-2e Ensembles Compare to ECMWF ENS on CRPS<\/h3>\n<p>EPT-2e, the ensemble variant of EPT-2, beats the 50-member ECMWF ENS mean on both RMSE and CRPS at virtually every lead time, as documented in arXiv:2507.09703. CRPS (continuous ranked probability score) is a proper scoring rule that rewards both accuracy and calibration, so a model that is accurate but overconfident scores worse than one that is accurate and well-calibrated. EPT-2e achieves this result with 10 published ensemble members against the ENS\u2019s 50 and updates 4 times per day operationally. No other AI weather model currently ships a productised ensemble equivalent. For energy traders who need probabilistic forecasts to size positions, manage imbalance risk, or price weather derivatives, EPT-2e\u2019s CRPS leadership over ECMWF ENS is the most directly relevant benchmark result in the 2025 StationBench evaluation.<\/p>\n<h2>Conclusion: EPT-2 and EPT-2e Set the 2025 Standard for Energy-Critical Forecasting<\/h2>\n<p>The 2025 StationBench results establish a clear hierarchy for AI weather model benchmarks on the variables that drive energy P&amp;L. EPT-2 leads ECMWF HRES, Microsoft Aurora, and GFS GraphCast on the four energy-critical variables, evaluated against more than 10,000 real ground stations with no post-processing. EPT-2e beats the 50-member ECMWF ENS mean on RMSE and CRPS at virtually every lead time. Both results are documented in peer-reviewed technical reports at <a href=\"https:\/\/arxiv.org\/abs\/2507.09703\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2507.09703<\/a> and <a href=\"https:\/\/arxiv.org\/abs\/2410.15076\" target=\"_blank\" rel=\"noindex nofollow\">arXiv:2410.15076<\/a>.<\/p>\n<p>The operational case is equally concrete. EPT-2 inference costs approximately $0.20\u2013$15 per run on a single GPU, which is roughly four orders of magnitude cheaper than a traditional NWP simulation. EPT-2 RR updates up to 24 times per day. A typical Jua run completes approximately 2.5 hours ahead of competing operational runs at the same cycle. A 1 GW wind portfolio that gains four percentage points of forecast accuracy saves approximately \u20ac1.5 M per year, and a 1 GW solar portfolio at the same accuracy gain saves approximately \u20ac3 M per year. Customers operating multi-GW portfolios scale these economics linearly, as <a href=\"https:\/\/nebius.com\/customer-stories\/jua\" target=\"_blank\">documented across Jua\u2019s customer base spanning utilities, trading houses, and quant funds across five continents<\/a>.<\/p>\n<p>Jua is a foundation model and agent company, and Jua for Energy is the first applied product. The Jua platform puts more than 25 models, including EPT-2, EPT-2e, ECMWF HRES, ECMWF ENS, Aurora, and GraphCast, on a single benchmarking surface. Any meteorologist, quant developer, or energy trader can run a live head-to-head comparison on their own region and variables in under five minutes, without a sales call.<\/p>\n<p><a href=\"https:\/\/meetings-eu1.hubspot.com\/guett\/energy-trading?uuid=d780665f-ff71-439c-addf-c80e49af0627\" target=\"_blank\"><strong>Run your own live benchmark in under five minutes<\/strong><\/a> on the Jua platform. The numbers speak.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Jua&#8217;s EPT-2 tops 2026 StationBench AI weather model benchmarks on wind, temperature &amp; solar \u2014 at 10,000x lower cost. See the full results.<\/p>\n","protected":false},"author":103,"featured_media":314,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"inline_featured_image":false,"footnotes":""},"categories":[11],"tags":[],"class_list":["post-315","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-weather-forecasting"],"_links":{"self":[{"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/posts\/315","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/comments?post=315"}],"version-history":[{"count":2,"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/posts\/315\/revisions"}],"predecessor-version":[{"id":784,"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/posts\/315\/revisions\/784"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/media\/314"}],"wp:attachment":[{"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/media?parent=315"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/categories?post=315"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/jua.ai\/articles\/wp-json\/wp\/v2\/tags?post=315"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}