{"id":17457,"date":"2026-01-10T11:07:45","date_gmt":"2026-01-10T11:07:45","guid":{"rendered":"https:\/\/www.quantumrun.com\/consulting\/?p=17457"},"modified":"2026-01-10T11:08:54","modified_gmt":"2026-01-10T11:08:54","slug":"perceiver-io-statistics","status":"publish","type":"post","link":"https:\/\/www.quantumrun.com\/consulting\/perceiver-io-statistics\/","title":{"rendered":"Perceiver IO Statistics [2026 Updated]"},"content":{"rendered":"\n<style>\n    \n    h2 {\n        font-size: 28px;\n        font-weight: 700;\n        margin: 40px 0 20px 0;\n        color: #1a1a1a;\n        line-height: 1.3;\n    }\n    \n    h3 {\n        font-size: 22px;\n        font-weight: 600;\n        margin: 30px 0 15px 0;\n        color: #2a2a2a;\n    }\n    \n    ul {\n        margin: 20px 0;\n        padding-left: 25px;\n    }\n    \n    li {\n        margin: 12px 0;\n        font-size: 16px;\n        line-height: 1.7;\n    }\n    \n    table {\n        width: 100%;\n        border-collapse: collapse;\n        margin: 25px 0;\n        font-size: 15px;\n        background: #fff;\n        box-shadow: 0 2px 8px rgba(0,0,0,0.08);\n    }\n    \n    th {\n        background: #f8f9fa;\n        padding: 14px;\n        text-align: left;\n        font-weight: 700;\n        border-bottom: 2px solid #e0e0e0;\n        color: #1a1a1a;\n    }\n    \n    td {\n        padding: 12px 14px;\n        border-bottom: 1px solid #f0f0f0;\n    }\n    \n    tr:hover {\n        background: #fafafa;\n    }\n    \n    .chart-container {\n        position: relative;\n        margin: 30px 0;\n        padding: 20px;\n        background: #fff;\n        border-radius: 8px;\n        box-shadow: 0 2px 12px rgba(0,0,0,0.06);\n height: 500px;\n    display: flex;\n    justify-content: center;\n    align-items: center;\n    }\n    \n    canvas {\n        max-width: 100%;\n        height: auto !important;\n    }\n    \n    details {\n        margin: 15px 0;\n        padding: 18px;\n        background: #fff;\n        border: 1px solid #e5e5e5;\n        border-radius: 6px;\n        cursor: pointer;\n        transition: all 0.3s ease;\n    }\n    \n    details:hover {\n        border-color: #d0d0d0;\n        box-shadow: 0 2px 8px rgba(0,0,0,0.05);\n    }\n    \n    details[open] {\n        border-color: #c0c0c0;\n    }\n    \n    summary {\n        font-weight: 600;\n        font-size: 17px;\n        color: #1a1a1a;\n        cursor: pointer;\n        user-select: none;\n        position: relative;\n        padding-left: 25px;\n    }\n    \n    summary::-webkit-details-marker {\n        display: none;\n    }\n    \n    summary::before {\n        \/\/content: '\u25b6';\n        position: absolute;\n        left: 0;\n        color: #666;\n        transition: transform 0.3s ease;\n    }\n    \n    details[open] summary::before {\n        transform: rotate(90deg);\n    }\n    \n    details p {\n        margin: 15px 0 5px 25px;\n        color: #555;\n        line-height: 1.7;\n    }\n    \n    .citations {\n        margin-top: 50px;\n        padding-top: 30px;\n        border-top: 2px solid #e5e5e5;\n    }\n    \n    .citations h2 {\n        font-size: 24px;\n        margin-bottom: 15px;\n    }\n    \n    .citations p {\n        margin: 10px 0;\n    }\n    \n    a {\n        color: #2563eb;\n        text-decoration: none;\n    }\n    \n    a:hover {\n        text-decoration: underline;\n    }\n    \n    @media (max-width: 768px) {\n        body {\n            padding: 15px;\n        }\n        \n        h2 {\n            font-size: 24px;\n        }\n        \n        h3 {\n            font-size: 20px;\n        }\n        \n        table {\n            font-size: 14px;\n        }\n        \n        th, td {\n            padding: 10px;\n        }\n    }\n<\/style>\n\n<p>Google DeepMind&#8217;s Perceiver IO achieved a 1.81 end-point error on Sintel Clean optical flow benchmarks, marking state-of-the-art performance without specialized flow architectures. The model processes 50,000+ pixels directly through attention mechanisms while maintaining linear computational scaling. Released in July 2021, Perceiver IO handles text, images, audio, and video through a unified architecture with 201 million parameters in its language variant.<\/p>\n\n<h2>Perceiver IO Key Statistics<\/h2>\n\n<ul>\n<li>Perceiver IO achieves 81.8 average GLUE score on multitask configuration, matching BERT baseline performance with 262-token vocabulary<\/li>\n<li>The architecture scales linearly O(MN) with input size compared to quadratic O(M\u00b2) scaling in standard Transformers<\/li>\n<li>Medical imaging applications recorded 99% accuracy for neurological disorder classification with 1.00 recall across six diseases<\/li>\n<li>AlphaStar integration demonstrated 3.5\u00d7 reduction in floating-point operations while maintaining 87% win rate<\/li>\n<li>Hugging Face hosts 36 Perceiver-related models as of 2024, including five primary pre-trained variants<\/li>\n<\/ul>\n\n<h2>Perceiver IO Model Architecture and Scale<\/h2>\n\n<p>The architecture employs 256 to 512 latent variables that process information independently of input and output dimensions. This design enables processing inputs with hundreds of thousands of dimensions while maintaining computational efficiency.<\/p>\n\n<p>Perceiver IO implements a read-process-write pipeline where inputs undergo cross-attention with latent variables, followed by self-attention in latent space, and conclude with cross-attention decoding for outputs.<\/p>\n\n<div class=\"chart-container\">\n    <canvas id=\"architectureChart\"><\/canvas>\n<\/div>\n\n<table>\n<tr>\n<th>Specification<\/th>\n<th>Value<\/th>\n<th>Notes<\/th>\n<\/tr>\n<tr>\n<td>Latent Variables<\/td>\n<td>256-512<\/td>\n<td>Configurable per task<\/td>\n<\/tr>\n<tr>\n<td>Input Scaling<\/td>\n<td>Linear O(MN)<\/td>\n<td>M=inputs, N=latents<\/td>\n<\/tr>\n<tr>\n<td>Output Scaling<\/td>\n<td>Linear O(ON)<\/td>\n<td>O=outputs, N=latents<\/td>\n<\/tr>\n<tr>\n<td>Latent Processing<\/td>\n<td>O(N\u00b2)<\/td>\n<td>Independent of I\/O size<\/td>\n<\/tr>\n<tr>\n<td>Max Input Pixels<\/td>\n<td>50,000+<\/td>\n<td>Direct pixel attention<\/td>\n<\/tr>\n<\/table>\n\n<h2>Perceiver IO Language Understanding Performance<\/h2>\n\n<p>The language model processes raw UTF-8 bytes without tokenization, achieving competitive results on natural language tasks. The architecture handles 2,048-byte input sequences with 26 processing layers.<\/p>\n\n<p>Perceiver IO&#8217;s multitask configuration recorded 81.8 GLUE score, exceeding BERT Base&#8217;s 81.1 baseline. The UTF-8 byte processing variant achieved 81.0, demonstrating minimal performance trade-off for eliminating tokenization requirements.<\/p>\n\n<div class=\"chart-container\">\n    <canvas id=\"languageChart\"><\/canvas>\n<\/div>\n\n<p>The vocabulary consists of 256 UTF-8 byte IDs plus 6 special tokens, totaling 262 tokens. This eliminates tokenizer training and vocabulary maintenance while processing approximately 201 million parameters.<\/p>\n\n<h2>Perceiver IO Computer Vision Benchmarks<\/h2>\n\n<p>ImageNet classification results reveal competitive performance without relying on 2D convolutional assumptions. The learned embeddings configuration achieved 72.7% top-1 accuracy despite having no prior knowledge of image 2D structure.<\/p>\n\n<p>Models with convolutional preprocessing reached 84.5% accuracy when pretrained on JFT dataset and 82.1% when pretrained on ImageNet alone. The architecture attends directly to pixels at 224\u00d7224 resolution.<\/p>\n\n<div class=\"chart-container\">\n    <canvas id=\"visionChart\"><\/canvas>\n<\/div>\n\n<h3>Vision Processing Specifications<\/h3>\n\n<p>The learned embeddings approach uses 1D position encoding and processes all pixels through direct attention. Convolutional preprocessing variants maintain compatibility with traditional vision pipelines while benefiting from Perceiver IO&#8217;s architectural advantages.<\/p>\n\n<h2>Perceiver IO Optical Flow Performance<\/h2>\n\n<p>Optical flow estimation represents the architecture&#8217;s most impressive achievement, surpassing PWCNet and RAFT on Sintel benchmarks. The model achieved 1.81 EPE on Sintel Clean and 2.42 EPE on Sintel Final.<\/p>\n\n<p>Training utilized AutoFlow, a synthetic dataset comprising 400,000 annotated image pairs at 368\u00d7496 resolution. The model concatenates image pairs along the channel dimension and extracts 3\u00d73 patches around each pixel, producing 54 values per pixel.<\/p>\n\n<div class=\"chart-container\">\n    <canvas id=\"opticalFlowChart\"><\/canvas>\n<\/div>\n\n<p>End-Point Error measures the average Euclidean distance between predicted and ground-truth flow vectors, with lower values indicating superior performance. Perceiver IO achieved these results without cost volumes, explicit warping, or hierarchical processing mechanisms.<\/p>\n\n<h2>Perceiver IO Multimodal Capabilities<\/h2>\n\n<p>Kinetics-700-2020 multimodal autoencoding demonstrates simultaneous processing of video, audio, and classification labels through shared latent representation. The configuration handles 16 frames at 224\u00d7224 resolution alongside 30,720 raw audio samples.<\/p>\n\n<table>\n<tr>\n<th>Metric<\/th>\n<th>Value<\/th>\n<th>Data Specification<\/th>\n<\/tr>\n<tr>\n<td>Video PSNR<\/td>\n<td>20.7 dB<\/td>\n<td>16 frames at 224\u00d7224<\/td>\n<\/tr>\n<tr>\n<td>Classification Top-1 Accuracy<\/td>\n<td>45%<\/td>\n<td>700 classes<\/td>\n<\/tr>\n<tr>\n<td>Video Patches Processed<\/td>\n<td>50,176<\/td>\n<td>4\u00d74 patch size<\/td>\n<\/tr>\n<tr>\n<td>Audio Samples Processed<\/td>\n<td>30,720<\/td>\n<td>Raw audio input<\/td>\n<\/tr>\n<tr>\n<td>Compression Ratio<\/td>\n<td>88\u00d7<\/td>\n<td>Latent bottleneck<\/td>\n<\/tr>\n<\/table>\n\n<p>The architecture processes inputs totaling 1,920 16-dimensional vectors from video and one 700-dimensional one-hot class label representation, demonstrating flexibility across diverse modalities.<\/p>\n\n<h2>Perceiver IO Medical Imaging Applications<\/h2>\n\n<p>A 2025 research study combining Vision Transformers with Perceiver IO demonstrated advances in multi-disease medical image classification. The framework achieved 1.00 recall for six diseases across neurological, dermatological, and pulmonological domains.<\/p>\n\n<p>This represents the first application of ViT + Perceiver IO hybrid architecture for these disease categories, outperforming conventional CNN architectures in accuracy, computational efficiency, and clinical interpretability.<\/p>\n\n<div class=\"chart-container\">\n    <canvas id=\"medicalChart\"><\/canvas>\n<\/div>\n\n<h3>Disease Classification Performance<\/h3>\n\n<p>Neurological disorders classification recorded 99% accuracy with 99% precision and 100% recall. Skin disease classification achieved 95% accuracy with 93% precision and 97% recall.<\/p>\n\n<p>Lung disease detection reached 98% accuracy with 97% precision and 100% recall, demonstrating consistent high performance across diverse medical imaging tasks.<\/p>\n\n<h2>Perceiver IO Computational Efficiency<\/h2>\n\n<p>The architecture&#8217;s fundamental advantage lies in decoupling processing complexity from input and output dimensions. Standard Transformers scale quadratically O(M\u00b2) with input size, while Perceiver IO maintains linear O(MN) scaling.<\/p>\n\n<div class=\"chart-container\">\n    <canvas id=\"efficiencyChart\"><\/canvas>\n<\/div>\n\n<p>BERT Base handles 512 tokens maximum, while Perceiver IO processes 2,048+ bytes without tokenization. The reduced latent size of 256 enables training networks with 26 processing layers compared to BERT&#8217;s 12 layers while maintaining similar FLOPs budgets.<\/p>\n\n<h3>StarCraft II Gaming Implementation<\/h3>\n\n<p>AlphaStar&#8217;s integration of Perceiver IO in entity processing modules achieved approximately 3.5\u00d7 reduction in floating-point operations. The architecture preserved parameter count while maintaining 87% win rate across experimental runs.<\/p>\n\n<table>\n<tr>\n<th>Metric<\/th>\n<th>Perceiver IO<\/th>\n<th>Original Transformer<\/th>\n<\/tr>\n<tr>\n<td>Win Rate<\/td>\n<td>87%<\/td>\n<td>87%<\/td>\n<\/tr>\n<tr>\n<td>FLOPs Reduction<\/td>\n<td>3.5\u00d7<\/td>\n<td>Baseline<\/td>\n<\/tr>\n<tr>\n<td>Experimental Runs<\/td>\n<td>3<\/td>\n<td>N\/A<\/td>\n<\/tr>\n<\/table>\n\n<h2>Perceiver IO Graph-Structured Data Extensions<\/h2>\n\n<p>Graph Perceiver IO (GPIO), published in Pattern Recognition in 2025, extends the architecture to graph-structured datasets while preserving multimodal capabilities. The model utilizes graph positional encoding and output query smoothing to handle topological information.<\/p>\n\n<p>GPIO demonstrates lower space complexity than traditional Graph Neural Networks with computation independent of adjacency matrix operations. Validation was conducted on RTX A6000, RTX 3090, and A100 GPUs using PyTorch 1.11.0 and PyTorch-Geometric 2.0.1.<\/p>\n\n<p>The architecture supports simultaneous processing of graph, text, and image data types, achieving competitive performance with state-of-the-art models on link prediction tasks.<\/p>\n\n<h2>Perceiver IO Availability and Implementation<\/h2>\n\n<p>Hugging Face Transformers integrated Perceiver IO on December 8, 2021, following the model&#8217;s initial release on July 30, 2021. Five primary pre-trained variants cover language modeling, image classification, optical flow estimation, and multimodal autoencoding.<\/p>\n\n<table>\n<tr>\n<th>Model Variant<\/th>\n<th>Task<\/th>\n<th>Primary Modality<\/th>\n<\/tr>\n<tr>\n<td>deepmind\/language-perceiver<\/td>\n<td>Masked Language Modeling<\/td>\n<td>Text (UTF-8 bytes)<\/td>\n<\/tr>\n<tr>\n<td>deepmind\/vision-perceiver-conv<\/td>\n<td>Image Classification<\/td>\n<td>Images (Conv preprocessing)<\/td>\n<\/tr>\n<tr>\n<td>deepmind\/vision-perceiver-learned<\/td>\n<td>Image Classification<\/td>\n<td>Images (Learned embeddings)<\/td>\n<\/tr>\n<tr>\n<td>deepmind\/optical-flow-perceiver<\/td>\n<td>Optical Flow Estimation<\/td>\n<td>Image Pairs<\/td>\n<\/tr>\n<tr>\n<td>deepmind\/multimodal-perceiver<\/td>\n<td>Autoencoding<\/td>\n<td>Video + Audio + Labels<\/td>\n<\/tr>\n<\/table>\n\n<p>As of 2024, 36 Perceiver-related models are available on the Hugging Face Model Hub, providing researchers and developers with pre-trained options across multiple domains and modalities.<\/p>\n\n<h2>FAQ<\/h2>\n\n<details>\n<summary>How many parameters does Perceiver IO have?<\/summary>\n<p>The Perceiver IO language variant contains approximately 201 million parameters. The model uses 256 latent variables with 26 processing layers and handles 2,048-byte input sequences while processing raw UTF-8 bytes without tokenization.<\/p>\n<\/details>\n\n<details>\n<summary>What makes Perceiver IO different from standard Transformers?<\/summary>\n<p>Perceiver IO employs self-attention on 256-512 fixed latent variables rather than inputs directly. This enables linear O(MN) scaling with input size instead of quadratic O(M\u00b2) scaling, allowing processing of hundreds of thousands of input dimensions efficiently.<\/p>\n<\/details>\n\n<details>\n<summary>What is Perceiver IO&#8217;s ImageNet accuracy?<\/summary>\n<p>Perceiver IO achieved 72.7% top-1 accuracy on ImageNet using only learned embeddings without 2D structural assumptions. Configurations with convolutional preprocessing reached 84.5% accuracy when pretrained on JFT dataset and 82.1% on ImageNet alone.<\/p>\n<\/details>\n\n<details>\n<summary>Can Perceiver IO process multiple data types simultaneously?<\/summary>\n<p>Yes, Perceiver IO handles text, images, audio, video, and combinations through unified architecture. The Kinetics-700 multimodal configuration simultaneously processes 16 video frames, 30,720 audio samples, and classification labels with 88\u00d7 compression ratio.<\/p>\n<\/details>\n\n<details>\n<summary>Where can I access pre-trained Perceiver IO models?<\/summary>\n<p>Hugging Face Transformers provides five pre-trained variants including language modeling, image classification, optical flow, and multimodal autoencoding. The library integrated Perceiver IO on December 8, 2021, with 36 related models currently available.<\/p>\n<\/details>\n\n<div class=\"citations\">\n<h2>Sources<\/h2>\n<p><a href=\"https:\/\/arxiv.org\/abs\/2107.14795\" rel=\"nofollow\">Perceiver IO: A General Architecture for Structured Inputs &#038; Outputs &#8211; arXiv<\/a><\/p>\n<p><a href=\"https:\/\/huggingface.co\/docs\/transformers\/model_doc\/perceiver\" rel=\"nofollow\" class=\"external\">Perceiver Documentation &#8211; Hugging Face<\/a><\/p>\n<p><a href=\"https:\/\/www.sciencedirect.com\/science\/article\/pii\/S0031320324008057\" rel=\"nofollow\">Graph Perceiver IO &#8211; Pattern Recognition<\/a><\/p>\n<p><a href=\"https:\/\/www.sciencedirect.com\/science\/article\/pii\/S1746809425004045\" rel=\"nofollow\">Medical Imaging with ViT and Perceiver IO &#8211; ScienceDirect<\/a><\/p>\n<\/div>\n\n<script src=\"https:\/\/cdn.jsdelivr.net\/npm\/chart.js\"><\/script>\n","protected":false},"excerpt":{"rendered":"<p>Google DeepMind&#8217;s Perceiver IO achieved a 1.81 end-point error on Sintel Clean optical flow benchmarks, marking state-of-the-art performance without specialized flow architectures. The model processes 50,000+ pixels directly through attention mechanisms while maintaining linear computational scaling. Released in July 2021, Perceiver IO handles text, images, audio, and video through a unified architecture with 201 million [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":17459,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"inline_featured_image":false,"_monsterinsights_skip_tracking":false,"footnotes":""},"categories":[130],"tags":[],"class_list":["post-17457","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.6 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Perceiver IO Statistics [2026 Updated]<\/title>\n<meta name=\"description\" content=\"Google DeepMind&#039;s Perceiver IO achieved a 1.81 end-point error on Sintel Clean optical flow benchmarks, marking state-of-the-art performance without\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.quantumrun.com\/consulting\/perceiver-io-statistics\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Perceiver IO Statistics [2026 Updated]\" \/>\n<meta property=\"og:description\" content=\"Google DeepMind&#039;s Perceiver IO achieved a 1.81 end-point error on Sintel Clean optical flow benchmarks, marking state-of-the-art performance without\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.quantumrun.com\/consulting\/perceiver-io-statistics\/\" \/>\n<meta property=\"og:site_name\" content=\"Quantumrun Foresight\" \/>\n<meta property=\"article:published_time\" content=\"2026-01-10T11:07:45+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-01-10T11:08:54+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.quantumrun.com\/consulting\/wp-content\/uploads\/2026\/01\/Perceiver-IO-Statistics-e1768043250228.png\" \/>\n\t<meta property=\"og:image:width\" content=\"1024\" \/>\n\t<meta property=\"og:image:height\" content=\"930\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"Richard Jaimes\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Richard Jaimes\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"6 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/perceiver-io-statistics\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/perceiver-io-statistics\\\/\"},\"author\":{\"name\":\"Richard Jaimes\",\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/#\\\/schema\\\/person\\\/5a0afcb410d8e3137f58d5a201809011\"},\"headline\":\"Perceiver IO Statistics [2026 Updated]\",\"datePublished\":\"2026-01-10T11:07:45+00:00\",\"dateModified\":\"2026-01-10T11:08:54+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/perceiver-io-statistics\\\/\"},\"wordCount\":1223,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/perceiver-io-statistics\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/wp-content\\\/uploads\\\/2026\\\/01\\\/Perceiver-IO-Statistics-e1768043250228.png\",\"articleSection\":[\"AI\"],\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/perceiver-io-statistics\\\/#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/perceiver-io-statistics\\\/\",\"url\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/perceiver-io-statistics\\\/\",\"name\":\"Perceiver IO Statistics [2026 Updated]\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/perceiver-io-statistics\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/perceiver-io-statistics\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/wp-content\\\/uploads\\\/2026\\\/01\\\/Perceiver-IO-Statistics-e1768043250228.png\",\"datePublished\":\"2026-01-10T11:07:45+00:00\",\"dateModified\":\"2026-01-10T11:08:54+00:00\",\"description\":\"Google DeepMind's Perceiver IO achieved a 1.81 end-point error on Sintel Clean optical flow benchmarks, marking state-of-the-art performance without\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/perceiver-io-statistics\\\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/perceiver-io-statistics\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/perceiver-io-statistics\\\/#primaryimage\",\"url\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/wp-content\\\/uploads\\\/2026\\\/01\\\/Perceiver-IO-Statistics-e1768043250228.png\",\"contentUrl\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/wp-content\\\/uploads\\\/2026\\\/01\\\/Perceiver-IO-Statistics-e1768043250228.png\",\"width\":1024,\"height\":930,\"caption\":\"Perceiver IO statistics showing 81.8 GLUE score, 1.81 optical flow EPE, and multimodal processing with linear computational scaling.\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/perceiver-io-statistics\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"AI\",\"item\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/ai\\\/\"},{\"@type\":\"ListItem\",\"position\":3,\"name\":\"Perceiver IO Statistics [2026 Updated]\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/#website\",\"url\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/\",\"name\":\"Quantumrun Foresight\",\"description\":\"Thrive from future trends.\",\"publisher\":{\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/#organization\",\"name\":\"Quantumrun Foresight\",\"url\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/wp-content\\\/uploads\\\/2020\\\/07\\\/cropped-con-logo-00.png\",\"contentUrl\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/wp-content\\\/uploads\\\/2020\\\/07\\\/cropped-con-logo-00.png\",\"width\":286,\"height\":55,\"caption\":\"Quantumrun Foresight\"},\"image\":{\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/#\\\/schema\\\/logo\\\/image\\\/\"}},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/#\\\/schema\\\/person\\\/5a0afcb410d8e3137f58d5a201809011\",\"name\":\"Richard Jaimes\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/552ad8a69a374ceb309e1045d345be2d8c30e8a0e8cda42acb1fc357ca381509?s=96&d=mm&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/552ad8a69a374ceb309e1045d345be2d8c30e8a0e8cda42acb1fc357ca381509?s=96&d=mm&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/552ad8a69a374ceb309e1045d345be2d8c30e8a0e8cda42acb1fc357ca381509?s=96&d=mm&r=g\",\"caption\":\"Richard Jaimes\"},\"sameAs\":[\"https:\\\/\\\/www.quantumrun.com\\\/consulting\"],\"url\":\"https:\\\/\\\/www.quantumrun.com\\\/consulting\\\/author\\\/admin-stcons\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Perceiver IO Statistics [2026 Updated]","description":"Google DeepMind's Perceiver IO achieved a 1.81 end-point error on Sintel Clean optical flow benchmarks, marking state-of-the-art performance without","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.quantumrun.com\/consulting\/perceiver-io-statistics\/","og_locale":"en_US","og_type":"article","og_title":"Perceiver IO Statistics [2026 Updated]","og_description":"Google DeepMind's Perceiver IO achieved a 1.81 end-point error on Sintel Clean optical flow benchmarks, marking state-of-the-art performance without","og_url":"https:\/\/www.quantumrun.com\/consulting\/perceiver-io-statistics\/","og_site_name":"Quantumrun Foresight","article_published_time":"2026-01-10T11:07:45+00:00","article_modified_time":"2026-01-10T11:08:54+00:00","og_image":[{"width":1024,"height":930,"url":"https:\/\/www.quantumrun.com\/consulting\/wp-content\/uploads\/2026\/01\/Perceiver-IO-Statistics-e1768043250228.png","type":"image\/png"}],"author":"Richard Jaimes","twitter_card":"summary_large_image","twitter_misc":{"Written by":"Richard Jaimes","Est. reading time":"6 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/www.quantumrun.com\/consulting\/perceiver-io-statistics\/#article","isPartOf":{"@id":"https:\/\/www.quantumrun.com\/consulting\/perceiver-io-statistics\/"},"author":{"name":"Richard Jaimes","@id":"https:\/\/www.quantumrun.com\/consulting\/#\/schema\/person\/5a0afcb410d8e3137f58d5a201809011"},"headline":"Perceiver IO Statistics [2026 Updated]","datePublished":"2026-01-10T11:07:45+00:00","dateModified":"2026-01-10T11:08:54+00:00","mainEntityOfPage":{"@id":"https:\/\/www.quantumrun.com\/consulting\/perceiver-io-statistics\/"},"wordCount":1223,"commentCount":0,"publisher":{"@id":"https:\/\/www.quantumrun.com\/consulting\/#organization"},"image":{"@id":"https:\/\/www.quantumrun.com\/consulting\/perceiver-io-statistics\/#primaryimage"},"thumbnailUrl":"https:\/\/www.quantumrun.com\/consulting\/wp-content\/uploads\/2026\/01\/Perceiver-IO-Statistics-e1768043250228.png","articleSection":["AI"],"inLanguage":"en-US","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/www.quantumrun.com\/consulting\/perceiver-io-statistics\/#respond"]}]},{"@type":"WebPage","@id":"https:\/\/www.quantumrun.com\/consulting\/perceiver-io-statistics\/","url":"https:\/\/www.quantumrun.com\/consulting\/perceiver-io-statistics\/","name":"Perceiver IO Statistics [2026 Updated]","isPartOf":{"@id":"https:\/\/www.quantumrun.com\/consulting\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.quantumrun.com\/consulting\/perceiver-io-statistics\/#primaryimage"},"image":{"@id":"https:\/\/www.quantumrun.com\/consulting\/perceiver-io-statistics\/#primaryimage"},"thumbnailUrl":"https:\/\/www.quantumrun.com\/consulting\/wp-content\/uploads\/2026\/01\/Perceiver-IO-Statistics-e1768043250228.png","datePublished":"2026-01-10T11:07:45+00:00","dateModified":"2026-01-10T11:08:54+00:00","description":"Google DeepMind's Perceiver IO achieved a 1.81 end-point error on Sintel Clean optical flow benchmarks, marking state-of-the-art performance without","breadcrumb":{"@id":"https:\/\/www.quantumrun.com\/consulting\/perceiver-io-statistics\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.quantumrun.com\/consulting\/perceiver-io-statistics\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.quantumrun.com\/consulting\/perceiver-io-statistics\/#primaryimage","url":"https:\/\/www.quantumrun.com\/consulting\/wp-content\/uploads\/2026\/01\/Perceiver-IO-Statistics-e1768043250228.png","contentUrl":"https:\/\/www.quantumrun.com\/consulting\/wp-content\/uploads\/2026\/01\/Perceiver-IO-Statistics-e1768043250228.png","width":1024,"height":930,"caption":"Perceiver IO statistics showing 81.8 GLUE score, 1.81 optical flow EPE, and multimodal processing with linear computational scaling."},{"@type":"BreadcrumbList","@id":"https:\/\/www.quantumrun.com\/consulting\/perceiver-io-statistics\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/www.quantumrun.com\/consulting\/"},{"@type":"ListItem","position":2,"name":"AI","item":"https:\/\/www.quantumrun.com\/consulting\/ai\/"},{"@type":"ListItem","position":3,"name":"Perceiver IO Statistics [2026 Updated]"}]},{"@type":"WebSite","@id":"https:\/\/www.quantumrun.com\/consulting\/#website","url":"https:\/\/www.quantumrun.com\/consulting\/","name":"Quantumrun Foresight","description":"Thrive from future trends.","publisher":{"@id":"https:\/\/www.quantumrun.com\/consulting\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.quantumrun.com\/consulting\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/www.quantumrun.com\/consulting\/#organization","name":"Quantumrun Foresight","url":"https:\/\/www.quantumrun.com\/consulting\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.quantumrun.com\/consulting\/#\/schema\/logo\/image\/","url":"https:\/\/www.quantumrun.com\/consulting\/wp-content\/uploads\/2020\/07\/cropped-con-logo-00.png","contentUrl":"https:\/\/www.quantumrun.com\/consulting\/wp-content\/uploads\/2020\/07\/cropped-con-logo-00.png","width":286,"height":55,"caption":"Quantumrun Foresight"},"image":{"@id":"https:\/\/www.quantumrun.com\/consulting\/#\/schema\/logo\/image\/"}},{"@type":"Person","@id":"https:\/\/www.quantumrun.com\/consulting\/#\/schema\/person\/5a0afcb410d8e3137f58d5a201809011","name":"Richard Jaimes","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/secure.gravatar.com\/avatar\/552ad8a69a374ceb309e1045d345be2d8c30e8a0e8cda42acb1fc357ca381509?s=96&d=mm&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/552ad8a69a374ceb309e1045d345be2d8c30e8a0e8cda42acb1fc357ca381509?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/552ad8a69a374ceb309e1045d345be2d8c30e8a0e8cda42acb1fc357ca381509?s=96&d=mm&r=g","caption":"Richard Jaimes"},"sameAs":["https:\/\/www.quantumrun.com\/consulting"],"url":"https:\/\/www.quantumrun.com\/consulting\/author\/admin-stcons\/"}]}},"_links":{"self":[{"href":"https:\/\/www.quantumrun.com\/consulting\/wp-json\/wp\/v2\/posts\/17457","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.quantumrun.com\/consulting\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.quantumrun.com\/consulting\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.quantumrun.com\/consulting\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.quantumrun.com\/consulting\/wp-json\/wp\/v2\/comments?post=17457"}],"version-history":[{"count":2,"href":"https:\/\/www.quantumrun.com\/consulting\/wp-json\/wp\/v2\/posts\/17457\/revisions"}],"predecessor-version":[{"id":17461,"href":"https:\/\/www.quantumrun.com\/consulting\/wp-json\/wp\/v2\/posts\/17457\/revisions\/17461"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.quantumrun.com\/consulting\/wp-json\/wp\/v2\/media\/17459"}],"wp:attachment":[{"href":"https:\/\/www.quantumrun.com\/consulting\/wp-json\/wp\/v2\/media?parent=17457"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.quantumrun.com\/consulting\/wp-json\/wp\/v2\/categories?post=17457"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.quantumrun.com\/consulting\/wp-json\/wp\/v2\/tags?post=17457"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}