{"id":39861,"date":"2026-10-22T17:26:00","date_gmt":"2026-10-22T17:26:00","guid":{"rendered":"https:\/\/www.tun.com\/home\/?p=39861"},"modified":"2026-10-22T20:23:59","modified_gmt":"2026-10-22T20:23:59","slug":"dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers","status":"publish","type":"post","link":"https:\/\/www.tun.com\/home\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\/","title":{"rendered":"Dharma AI&#8217;s VocabX Cuts Inference Cost 77% by Fixing Tokenizers"},"content":{"rendered":"\n<div class=\"wp-block-group\"><div class=\"wp-block-group__inner-container is-layout-constrained wp-block-group-is-layout-constrained\">\n<div class=\"wp-block-uagb-blockquote uagb-block-e7eb3fc3 uagb-blockquote__skin-border uagb-blockquote__stack-img-none\"><blockquote class=\"uagb-blockquote\"><div class=\"uagb-blockquote__content\">Researchers at Dharma AI found that the tokenizer \u2014 not the model \u2014 is one of the biggest cost levers in specialized AI applications. Their VocabX method cut measured inference cost by 77% in a production essay-grading system without changing the underlying model at all.<\/div><footer><div class=\"uagb-blockquote__author-wrap uagb-blockquote__author-at-left\"><\/div><\/footer><\/blockquote><\/div>\n\n\n\n<div class=\"wp-block-group is-content-justification-space-between is-nowrap is-layout-flex wp-container-core-group-is-layout-b0ffac9c wp-block-group-is-layout-flex\"><div style=\"font-size:16px\" class=\"has-text-align-left wp-block-post-author\"><div class=\"wp-block-post-author__content\"><p class=\"wp-block-post-author__name\">The University Network<\/p><\/div><\/div>\n\n\n<div class=\"wp-block-uagb-social-share uagb-social-share__outer-wrap uagb-social-share__layout-horizontal uagb-block-ee584a31\">\n<div class=\"wp-block-uagb-social-share-child uagb-ss-repeater uagb-ss__wrapper uagb-block-ec619ce7\"><span class=\"uagb-ss__link\" data-href=\"https:\/\/www.facebook.com\/sharer.php?u=\" tabindex=\"0\" role=\"button\" aria-label=\"facebook\"><span class=\"uagb-ss__source-wrap\"><span class=\"uagb-ss__source-icon\"><svg xmlns=\"https:\/\/www.w3.org\/2000\/svg\" viewBox=\"0 0 512 512\"><path d=\"M504 256C504 119 393 8 256 8S8 119 8 256c0 123.8 90.69 226.4 209.3 245V327.7h-63V256h63v-54.64c0-62.15 37-96.48 93.67-96.48 27.14 0 55.52 4.84 55.52 4.84v61h-31.28c-30.8 0-40.41 19.12-40.41 38.73V256h68.78l-11 71.69h-57.78V501C413.3 482.4 504 379.8 504 256z\"><\/path><\/svg><\/span><\/span><\/span><\/div>\n\n\n\n<div class=\"wp-block-uagb-social-share-child uagb-ss-repeater uagb-ss__wrapper uagb-block-32d99934\"><span class=\"uagb-ss__link\" data-href=\"https:\/\/twitter.com\/share?url=\" tabindex=\"0\" role=\"button\" aria-label=\"twitter\"><span class=\"uagb-ss__source-wrap\"><span class=\"uagb-ss__source-icon\"><svg xmlns=\"https:\/\/www.w3.org\/2000\/svg\" viewBox=\"0 0 512 512\"><path d=\"M389.2 48h70.6L305.6 224.2 487 464H345L233.7 318.6 106.5 464H35.8L200.7 275.5 26.8 48H172.4L272.9 180.9 389.2 48zM364.4 421.8h39.1L151.1 88h-42L364.4 421.8z\"><\/path><\/svg><\/span><\/span><\/span><\/div>\n\n\n\n<div class=\"wp-block-uagb-social-share-child uagb-ss-repeater uagb-ss__wrapper uagb-block-1d136f14\"><span class=\"uagb-ss__link\" data-href=\"https:\/\/www.linkedin.com\/shareArticle?url=\" tabindex=\"0\" role=\"button\" aria-label=\"linkedin\"><span class=\"uagb-ss__source-wrap\"><span class=\"uagb-ss__source-icon\"><svg xmlns=\"https:\/\/www.w3.org\/2000\/svg\" viewBox=\"0 0 448 512\"><path d=\"M416 32H31.9C14.3 32 0 46.5 0 64.3v383.4C0 465.5 14.3 480 31.9 480H416c17.6 0 32-14.5 32-32.3V64.3c0-17.8-14.4-32.3-32-32.3zM135.4 416H69V202.2h66.5V416zm-33.2-243c-21.3 0-38.5-17.3-38.5-38.5S80.9 96 102.2 96c21.2 0 38.5 17.3 38.5 38.5 0 21.3-17.2 38.5-38.5 38.5zm282.1 243h-66.4V312c0-24.8-.5-56.7-34.5-56.7-34.6 0-39.9 27-39.9 54.9V416h-66.4V202.2h63.7v29.2h.9c8.9-16.8 30.6-34.5 62.9-34.5 67.2 0 79.7 44.3 79.7 101.9V416z\"><\/path><\/svg><\/span><\/span><\/span><\/div>\n<\/div>\n<\/div>\n<\/div><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">When engineers at Dharma AI looked at their production Portuguese essay-grading system \u2014 built on Qwen3-4B, a 4-billion-parameter model already trimmed for efficiency \u2014 they found a surprisingly large cost leak hiding in plain sight. Not in the model weights. Not in the hardware. In the tokenizer.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The team has previewed results from an upcoming paper introducing a method called VocabX, which specializes a model&#8217;s vocabulary around the actual dollar cost of generating tokens rather than how frequently tokens appear. Applied to their essay grader, which receives a Portuguese essay and returns a JSON assessment with a grade and comment for each competency, the approach cut measured inference cost by 77% \u2014 without retraining the base model, changing the training data or switching hardware.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Why the Tokenizer Is a Hidden Cost Driver<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Most developers who fine-tune a general-purpose model like Qwen, Gemma or Llama inherit its tokenizer unchanged and move on. That tokenizer was built to handle everything: source code, legal text, Chinese, Portuguese, JSON, markdown. For a general-purpose model, that breadth is a feature. For a model doing exactly one job, it creates two kinds of waste.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">First, fragmentation: the tokenizer splits the phrases a specialized model writes most often into far more pieces than necessary. In the essay grader, every response opens with the same JSON envelope, the same competency names, the same evaluative phrases \u2014 sequences that cost several tokens each when one would do. Second, excess: the vocabulary table contains tens of thousands of entries the task will almost never use, and the model scores every single one of them at every generation step.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">That second problem is more expensive than it looks, and the reason comes down to how inference actually works. During the input-processing phase, called prefill, the model handles all input tokens in one pass, so GPU memory reads are shared efficiently across tokens. During the output-generation phase, called decode, the model produces one token at a time, and each step requires a full pass through the network. In that regime, the GPU spends most of its time waiting on memory bandwidth, not computing. In the Dharma AI grader specifically, generating one token costs 7.4 times as much as reading one input token.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">That asymmetry is the core insight VocabX exploits. Prior vocabulary-adaptation methods \u2014 including fast vocabulary transfer (FVT), multi-word tokenization (MWT) and Liquid AI&#8217;s in-place BPE \u2014 reduce fragmentation but optimize for token frequency, treating every token as equally valuable regardless of whether it appears in the input or the output. A method that compresses both sides equally misses most of the available savings when output tokens are seven times more expensive.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">What VocabX Actually Does Differently<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">VocabX makes two structural changes. First, it reframes the selection criterion: instead of asking how many tokens a candidate vocabulary entry saves, it asks how much cost it saves, weighting each saved token by where in the conversation it appears. Second, it operates on two vocabulary layers simultaneously. At the sub-word level it rebuilds how domain words are segmented. At a higher level it identifies long repeated spans \u2014 full JSON envelopes, standard evaluation phrases, entire lines \u2014 and collapses each into a single token the paper calls a macro token. Both layers are optimized under the same cost criterion.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">To find these repeating spans, VocabX uses five specialized miners targeting different repetition types: long words, whole lines, short expressions, complete sentences and JSON or tag structures. Removing any one of those miners, according to the paper, would drop token reduction on the grader by 55.6 percentage points \u2014 a sign of how diverse the waste actually is in a production system.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The result was a vocabulary table that shrank from roughly 152,000 entries to roughly 65,000, reducing the size of the model&#8217;s output layer \u2014 which alone accounts for nearly 10% of Qwen3-4B&#8217;s parameters \u2014 and cutting the memory bandwidth required at every decode step.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">How It Compares to Existing Methods<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Tested across five domains \u2014 contracts, medical papers, scientific papers, news and Python code \u2014 and three base models (Qwen3, Llama-3.1 and Gemma-3), VocabX led all 15 domain-model combinations against FVT, MWT and in-place BPE on estimated inference-cost reduction. The margins were 10.2 percentage points over the next-best method on Qwen3, 7.5 on Llama-3.1, and 13.6 on Gemma-3. For context, FVT has been shown to achieve roughly a 32% reduction in tokens per sequence in highly specialized domains like medicine \u2014 meaningful, but well short of VocabX&#8217;s figures, and without cost-weighting.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Important caveats apply. The paper is still under review and results may change before publication. The 77% figure comes from one system on one hardware configuration; independent replication across domains will be the real test. All versions in the study ran on a single RTX 5090 GPU using the open-source vLLM inference server.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">What This Means If You&#8217;re Building AI Systems<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">For students and early-career developers, the most common deployment pattern is exactly what VocabX targets: fine-tune a general-purpose model, inherit the full tokenizer, deploy and then absorb the cloud bill. The VocabX approach requires no new training data and no new hardware \u2014 the only variable across all tested configurations was the tokenizer.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Even before the full VocabX codebase is released pending peer review, the conceptual framework is immediately usable. Whether you&#8217;re building a legal clause extractor, a medical triage assistant, a code reviewer or any other single-domain application, the design principle transfers directly: profile what your model writes most, weight token savings by generation cost rather than frequency, and treat the tokenizer as an optimization target rather than a fixed inheritance. In a job interview or a capstone review, being able to articulate tokenizer economics \u2014 and why output tokens are more expensive than input tokens in memory-bound inference \u2014 signals production-systems thinking that most new graduates haven&#8217;t developed yet.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">VocabX fits into a broader industry push toward smaller, cheaper, purpose-built models. NVIDIA research has argued that small language models are now capable enough for the vast majority of agentic subtasks and can reduce costs by 10 to 30 times compared to frontier models. VocabX goes one level further: it squeezes additional efficiency out of small models that have already been chosen for cost reasons, without requiring a different base model or better hardware.<\/p>\n\n\n\n<div style=\"height:3px\" aria-hidden=\"true\" class=\"wp-block-spacer\"><\/div>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Source<\/strong>: <a href=\"https:\/\/huggingface.co\/blog\/Dharma-AI\/vocabx-vocabulary-specialization\" target=\"_blank\" rel=\"noopener\">Dharma AI via Hugging Face<\/a><\/p>\n\n\n\n<details class=\"research-citations\">\n<summary>Additional research sources<\/summary>\n<ul>\n<li><a href=\"https:\/\/hyper.ai\/en\/stories\/47ba5de6cc8455014d0c0d80ea3d2acf\" rel=\"nofollow noopener\" target=\"_blank\">https:\/\/hyper.ai\/en\/stories\/47ba5de6cc8455014d0c0d80ea3d2acf<\/a><\/li>\n<li><a href=\"https:\/\/superpowerdaily.com\/posts\/dharma-ai-says-a-tailored-ai-vocabulary-cut-essay-grading-inference-cost-77\" rel=\"nofollow noopener\" target=\"_blank\">https:\/\/superpowerdaily.com\/posts\/dharma-ai-says-a-tailored-ai-vocabulary-cut-essay-grading-inference-cost-77<\/a><\/li>\n<li><a href=\"https:\/\/daily.dev\/posts\/how-specializing-the-vocabulary-of-slms-can-cut-inference-cost-by-77-in-real-applications-a4vqrjdnh\" rel=\"nofollow noopener\" target=\"_blank\">https:\/\/daily.dev\/posts\/how-specializing-the-vocabulary-of-slms-can-cut-inference-cost-by-77-in-real-applications-a4vqrjdnh<\/a><\/li>\n<li><a href=\"https:\/\/runtimewire.com\/article\/dharma-ai-vocabx-inference-cost\" rel=\"nofollow noopener\" target=\"_blank\">https:\/\/runtimewire.com\/article\/dharma-ai-vocabx-inference-cost<\/a><\/li>\n<li><a href=\"https:\/\/www.liquid.ai\/blog\/lfm2-5-8b-a1b\" rel=\"nofollow noopener\" target=\"_blank\">https:\/\/www.liquid.ai\/blog\/lfm2-5-8b-a1b<\/a><\/li>\n<\/ul>\n<\/details>\n","protected":false},"excerpt":{"rendered":"<p>Researchers at Dharma AI found that the tokenizer \u2014 not the model \u2014 is one of the biggest cost levers in specialized AI applications. Their VocabX method cut measured inference cost by 77% in a production essay-grading system without changing the underlying model at all.<\/p>\n","protected":false},"author":3,"featured_media":39860,"comment_status":"open","ping_status":"open","sticky":false,"template":"single-no-separators","format":"standard","meta":{"_acf_changed":false,"_uag_custom_page_level_css":"","_monsterinsights_skip_tracking":false,"footnotes":""},"categories":[8],"tags":[4082,3346,3954,3637,4081,4080,3347,3683,4079],"class_list":["post-39861","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai","tag-dharma-ai","tag-hugging-face","tag-inference-optimization-2","tag-liquid-ai","tag-mlops","tag-nlp","tag-nvidia","tag-small-language-models","tag-tokenization"],"acf":[],"aioseo_notices":[],"aioseo_head":"\n\t\t<!-- All in One SEO 5.0.3 - aioseo.com -->\n\t<meta name=\"description\" content=\"Researchers at Dharma AI found that the tokenizer \u2014 not the model \u2014 is one of the biggest cost levers in specialized AI applications. Their VocabX method cut measured inference cost by 77% in a production essay-grading system without changing the underlying model at all.\" \/>\n\t<meta name=\"robots\" content=\"max-image-preview:large\" \/>\n\t<meta name=\"author\" content=\"The University Network\"\/>\n\t<link rel=\"canonical\" href=\"https:\/\/www.tun.com\/home\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\/\" \/>\n\t\t<meta name=\"generator\" content=\"All in One SEO (AIOSEO) 5.0.3\" \/>\n\t\t<meta property=\"og:locale\" content=\"en_US\" \/>\n\t\t<meta property=\"og:site_name\" content=\"TUN - The University Network\" \/>\n\t\t<meta property=\"og:type\" content=\"article\" \/>\n\t\t<meta property=\"og:title\" content=\"Dharma AI\u2019s VocabX Cuts Inference Cost 77% by Fixing Tokenizers - TUN\" \/>\n\t\t<meta property=\"og:description\" content=\"Researchers at Dharma AI found that the tokenizer \u2014 not the model \u2014 is one of the biggest cost levers in specialized AI applications. Their VocabX method cut measured inference cost by 77% in a production essay-grading system without changing the underlying model at all.\" \/>\n\t\t<meta property=\"og:url\" content=\"https:\/\/www.tun.com\/home\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\/\" \/>\n\t\t<meta property=\"article:published_time\" content=\"2026-10-22T17:26:00+00:00\" \/>\n\t\t<meta property=\"article:modified_time\" content=\"2026-10-22T20:23:59+00:00\" \/>\n\t\t<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n\t\t<meta name=\"twitter:title\" content=\"Dharma AI\u2019s VocabX Cuts Inference Cost 77% by Fixing Tokenizers - TUN\" \/>\n\t\t<meta name=\"twitter:description\" content=\"Researchers at Dharma AI found that the tokenizer \u2014 not the model \u2014 is one of the biggest cost levers in specialized AI applications. Their VocabX method cut measured inference cost by 77% in a production essay-grading system without changing the underlying model at all.\" \/>\n\t\t<script type=\"application\/ld+json\" class=\"aioseo-schema\">\n\t\t\t{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"BlogPosting\",\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\\\/#blogposting\",\"name\":\"Dharma AI\\u2019s VocabX Cuts Inference Cost 77% by Fixing Tokenizers - TUN\",\"headline\":\"Dharma AI&#8217;s VocabX Cuts Inference Cost 77% by Fixing Tokenizers\",\"author\":{\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/author\\\/funky_junkie\\\/#author\"},\"publisher\":{\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/#organization\"},\"image\":{\"@type\":\"ImageObject\",\"url\":\"https:\\\/\\\/www.tun.com\\\/home\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizer.png\",\"width\":1536,\"height\":1024,\"caption\":\"Abstract grid of tokens compressing into a smaller specialized vocabulary structure\"},\"datePublished\":\"2026-10-22T17:26:00+00:00\",\"dateModified\":\"2026-10-22T20:23:59+00:00\",\"inLanguage\":\"en-US\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\\\/#webpage\"},\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\\\/#webpage\"},\"articleSection\":\"AI, Dharma AI, Hugging Face, inference optimization, Liquid AI, mlops, nlp, NVIDIA, small language models, tokenization\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\\\/#breadcrumblist\",\"itemListElement\":[{\"@type\":\"ListItem\",\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/#listItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/www.tun.com\\\/home\\\/\",\"nextItem\":{\"@type\":\"ListItem\",\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/category\\\/ai\\\/#listItem\",\"name\":\"AI\"}},{\"@type\":\"ListItem\",\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/category\\\/ai\\\/#listItem\",\"position\":2,\"name\":\"AI\",\"item\":\"https:\\\/\\\/www.tun.com\\\/home\\\/category\\\/ai\\\/\",\"nextItem\":{\"@type\":\"ListItem\",\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\\\/#listItem\",\"name\":\"Dharma AI&#8217;s VocabX Cuts Inference Cost 77% by Fixing Tokenizers\"},\"previousItem\":{\"@type\":\"ListItem\",\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/#listItem\",\"name\":\"Home\"}},{\"@type\":\"ListItem\",\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\\\/#listItem\",\"position\":3,\"name\":\"Dharma AI&#8217;s VocabX Cuts Inference Cost 77% by Fixing Tokenizers\",\"previousItem\":{\"@type\":\"ListItem\",\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/category\\\/ai\\\/#listItem\",\"name\":\"AI\"}}]},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/#organization\",\"name\":\"Home\",\"description\":\"The University Network\",\"url\":\"https:\\\/\\\/www.tun.com\\\/home\\\/\"},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/author\\\/funky_junkie\\\/#author\",\"url\":\"https:\\\/\\\/www.tun.com\\\/home\\\/author\\\/funky_junkie\\\/\",\"name\":\"The University Network\",\"image\":{\"@type\":\"ImageObject\",\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\\\/#authorImage\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/13dd0893ea3180b8e23da81ab702e21a881401f7778d2b4aedbd75103e0b7e4f?s=96&d=mm&r=g\",\"width\":96,\"height\":96,\"caption\":\"The University Network\"}},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\\\/#webpage\",\"url\":\"https:\\\/\\\/www.tun.com\\\/home\\\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\\\/\",\"name\":\"Dharma AI\\u2019s VocabX Cuts Inference Cost 77% by Fixing Tokenizers - TUN\",\"description\":\"Researchers at Dharma AI found that the tokenizer \\u2014 not the model \\u2014 is one of the biggest cost levers in specialized AI applications. Their VocabX method cut measured inference cost by 77% in a production essay-grading system without changing the underlying model at all.\",\"inLanguage\":\"en-US\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/#website\"},\"breadcrumb\":{\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\\\/#breadcrumblist\"},\"author\":{\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/author\\\/funky_junkie\\\/#author\"},\"creator\":{\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/author\\\/funky_junkie\\\/#author\"},\"image\":{\"@type\":\"ImageObject\",\"url\":\"https:\\\/\\\/www.tun.com\\\/home\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizer.png\",\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\\\/#mainImage\",\"width\":1536,\"height\":1024,\"caption\":\"Abstract grid of tokens compressing into a smaller specialized vocabulary structure\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\\\/#mainImage\"},\"datePublished\":\"2026-10-22T17:26:00+00:00\",\"dateModified\":\"2026-10-22T20:23:59+00:00\"},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/#website\",\"url\":\"https:\\\/\\\/www.tun.com\\\/home\\\/\",\"name\":\"Home\",\"description\":\"The University Network\",\"inLanguage\":\"en-US\",\"publisher\":{\"@id\":\"https:\\\/\\\/www.tun.com\\\/home\\\/#organization\"}}]}\n\t\t<\/script>\n\t\t<!-- All in One SEO -->\n\n","aioseo_head_json":{"title":"Dharma AI\u2019s VocabX Cuts Inference Cost 77% by Fixing Tokenizers - TUN","description":"Researchers at Dharma AI found that the tokenizer \u2014 not the model \u2014 is one of the biggest cost levers in specialized AI applications. Their VocabX method cut measured inference cost by 77% in a production essay-grading system without changing the underlying model at all.","canonical_url":"https:\/\/www.tun.com\/home\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\/","robots":"max-image-preview:large","keywords":"","webmasterTools":{"miscellaneous":""},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"BlogPosting","@id":"https:\/\/www.tun.com\/home\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\/#blogposting","name":"Dharma AI\u2019s VocabX Cuts Inference Cost 77% by Fixing Tokenizers - TUN","headline":"Dharma AI&#8217;s VocabX Cuts Inference Cost 77% by Fixing Tokenizers","author":{"@id":"https:\/\/www.tun.com\/home\/author\/funky_junkie\/#author"},"publisher":{"@id":"https:\/\/www.tun.com\/home\/#organization"},"image":{"@type":"ImageObject","url":"https:\/\/www.tun.com\/home\/wp-content\/uploads\/2026\/10\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizer.png","width":1536,"height":1024,"caption":"Abstract grid of tokens compressing into a smaller specialized vocabulary structure"},"datePublished":"2026-10-22T17:26:00+00:00","dateModified":"2026-10-22T20:23:59+00:00","inLanguage":"en-US","mainEntityOfPage":{"@id":"https:\/\/www.tun.com\/home\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\/#webpage"},"isPartOf":{"@id":"https:\/\/www.tun.com\/home\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\/#webpage"},"articleSection":"AI, Dharma AI, Hugging Face, inference optimization, Liquid AI, mlops, nlp, NVIDIA, small language models, tokenization"},{"@type":"BreadcrumbList","@id":"https:\/\/www.tun.com\/home\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\/#breadcrumblist","itemListElement":[{"@type":"ListItem","@id":"https:\/\/www.tun.com\/home\/#listItem","position":1,"name":"Home","item":"https:\/\/www.tun.com\/home\/","nextItem":{"@type":"ListItem","@id":"https:\/\/www.tun.com\/home\/category\/ai\/#listItem","name":"AI"}},{"@type":"ListItem","@id":"https:\/\/www.tun.com\/home\/category\/ai\/#listItem","position":2,"name":"AI","item":"https:\/\/www.tun.com\/home\/category\/ai\/","nextItem":{"@type":"ListItem","@id":"https:\/\/www.tun.com\/home\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\/#listItem","name":"Dharma AI&#8217;s VocabX Cuts Inference Cost 77% by Fixing Tokenizers"},"previousItem":{"@type":"ListItem","@id":"https:\/\/www.tun.com\/home\/#listItem","name":"Home"}},{"@type":"ListItem","@id":"https:\/\/www.tun.com\/home\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\/#listItem","position":3,"name":"Dharma AI&#8217;s VocabX Cuts Inference Cost 77% by Fixing Tokenizers","previousItem":{"@type":"ListItem","@id":"https:\/\/www.tun.com\/home\/category\/ai\/#listItem","name":"AI"}}]},{"@type":"Organization","@id":"https:\/\/www.tun.com\/home\/#organization","name":"Home","description":"The University Network","url":"https:\/\/www.tun.com\/home\/"},{"@type":"Person","@id":"https:\/\/www.tun.com\/home\/author\/funky_junkie\/#author","url":"https:\/\/www.tun.com\/home\/author\/funky_junkie\/","name":"The University Network","image":{"@type":"ImageObject","@id":"https:\/\/www.tun.com\/home\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\/#authorImage","url":"https:\/\/secure.gravatar.com\/avatar\/13dd0893ea3180b8e23da81ab702e21a881401f7778d2b4aedbd75103e0b7e4f?s=96&d=mm&r=g","width":96,"height":96,"caption":"The University Network"}},{"@type":"WebPage","@id":"https:\/\/www.tun.com\/home\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\/#webpage","url":"https:\/\/www.tun.com\/home\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\/","name":"Dharma AI\u2019s VocabX Cuts Inference Cost 77% by Fixing Tokenizers - TUN","description":"Researchers at Dharma AI found that the tokenizer \u2014 not the model \u2014 is one of the biggest cost levers in specialized AI applications. Their VocabX method cut measured inference cost by 77% in a production essay-grading system without changing the underlying model at all.","inLanguage":"en-US","isPartOf":{"@id":"https:\/\/www.tun.com\/home\/#website"},"breadcrumb":{"@id":"https:\/\/www.tun.com\/home\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\/#breadcrumblist"},"author":{"@id":"https:\/\/www.tun.com\/home\/author\/funky_junkie\/#author"},"creator":{"@id":"https:\/\/www.tun.com\/home\/author\/funky_junkie\/#author"},"image":{"@type":"ImageObject","url":"https:\/\/www.tun.com\/home\/wp-content\/uploads\/2026\/10\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizer.png","@id":"https:\/\/www.tun.com\/home\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\/#mainImage","width":1536,"height":1024,"caption":"Abstract grid of tokens compressing into a smaller specialized vocabulary structure"},"primaryImageOfPage":{"@id":"https:\/\/www.tun.com\/home\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\/#mainImage"},"datePublished":"2026-10-22T17:26:00+00:00","dateModified":"2026-10-22T20:23:59+00:00"},{"@type":"WebSite","@id":"https:\/\/www.tun.com\/home\/#website","url":"https:\/\/www.tun.com\/home\/","name":"Home","description":"The University Network","inLanguage":"en-US","publisher":{"@id":"https:\/\/www.tun.com\/home\/#organization"}}]},"og:locale":"en_US","og:site_name":"TUN - The University Network","og:type":"article","og:title":"Dharma AI\u2019s VocabX Cuts Inference Cost 77% by Fixing Tokenizers - TUN","og:description":"Researchers at Dharma AI found that the tokenizer \u2014 not the model \u2014 is one of the biggest cost levers in specialized AI applications. Their VocabX method cut measured inference cost by 77% in a production essay-grading system without changing the underlying model at all.","og:url":"https:\/\/www.tun.com\/home\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\/","article:published_time":"2026-10-22T17:26:00+00:00","article:modified_time":"2026-10-22T20:23:59+00:00","twitter:card":"summary_large_image","twitter:title":"Dharma AI\u2019s VocabX Cuts Inference Cost 77% by Fixing Tokenizers - TUN","twitter:description":"Researchers at Dharma AI found that the tokenizer \u2014 not the model \u2014 is one of the biggest cost levers in specialized AI applications. Their VocabX method cut measured inference cost by 77% in a production essay-grading system without changing the underlying model at all."},"aioseo_meta_data":{"post_id":"39861","title":null,"description":null,"keywords":null,"keyphrases":{"focus":{"keyphrase":"","score":0,"analysis":{"keyphraseInTitle":{"score":0,"maxScore":9,"error":1}}},"additional":[]},"primary_term":null,"canonical_url":null,"og_title":null,"og_description":null,"og_object_type":"default","og_image_type":"default","og_image_url":null,"og_image_width":null,"og_image_height":null,"og_image_custom_url":null,"og_image_custom_fields":null,"og_video":"","og_custom_url":null,"og_article_section":null,"og_article_tags":null,"twitter_use_og":false,"twitter_card":"default","twitter_image_type":"default","twitter_image_url":null,"twitter_image_custom_url":null,"twitter_image_custom_fields":null,"twitter_title":null,"twitter_description":null,"schema":{"blockGraphs":[],"customGraphs":[],"default":{"data":{"Article":[],"Course":[],"Dataset":[],"FAQPage":[],"Movie":[],"Person":[],"Product":[],"ProductReview":[],"Car":[],"Recipe":[],"Service":[],"SoftwareApplication":[],"WebPage":[]},"graphName":"BlogPosting","isEnabled":true},"graphs":[]},"schema_type":"default","schema_type_options":null,"pillar_content":false,"robots_default":true,"robots_noindex":false,"robots_noarchive":false,"robots_nosnippet":false,"robots_nofollow":false,"robots_noimageindex":false,"robots_noodp":false,"robots_notranslate":false,"robots_max_snippet":"-1","robots_max_videopreview":"-1","robots_max_imagepreview":"large","priority":null,"frequency":"default","local_seo":null,"breadcrumb_settings":null,"limit_modified_date":true,"ai":{"faqs":[],"keyPoints":[],"schemas":[],"titles":[],"descriptions":[],"socialPosts":{"email":{"subject":"","preview":"","content":""},"linkedin":[],"twitter":[],"facebook":[],"instagram":[]}},"created":"2026-10-22 15:28:04","updated":"2026-10-22 20:51:43","seo_analyzer_scan_date":null,"focus_keyword":null,"additional_keywords":null,"truseo_locale":null},"aioseo_breadcrumb":"<div class=\"aioseo-breadcrumbs\"><span class=\"aioseo-breadcrumb\">\n\t\t\t<a href=\"https:\/\/www.tun.com\/home\/\" title=\"Home\">Home<\/a>\n\t\t<\/span><span class=\"aioseo-breadcrumb-separator\">\u00bb<\/span><span class=\"aioseo-breadcrumb\">\n\t\t\t<a href=\"https:\/\/www.tun.com\/home\/category\/ai\/\" title=\"AI\">AI<\/a>\n\t\t<\/span><span class=\"aioseo-breadcrumb-separator\">\u00bb<\/span><span class=\"aioseo-breadcrumb\">\n\t\t\tDharma AI\u2019s VocabX Cuts Inference Cost 77% by Fixing Tokenizers\n\t\t<\/span><\/div>","aioseo_breadcrumb_json":[{"label":"Home","link":"https:\/\/www.tun.com\/home\/"},{"label":"AI","link":"https:\/\/www.tun.com\/home\/category\/ai\/"},{"label":"Dharma AI&#8217;s VocabX Cuts Inference Cost 77% by Fixing Tokenizers","link":"https:\/\/www.tun.com\/home\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizers\/"}],"uagb_featured_image_src":{"full":["https:\/\/www.tun.com\/home\/wp-content\/uploads\/2026\/10\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizer.png",1536,1024,false],"thumbnail":["https:\/\/www.tun.com\/home\/wp-content\/uploads\/2026\/10\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizer-150x150.png",150,150,true],"medium":["https:\/\/www.tun.com\/home\/wp-content\/uploads\/2026\/10\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizer-300x200.png",300,200,true],"medium_large":["https:\/\/www.tun.com\/home\/wp-content\/uploads\/2026\/10\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizer-768x512.png",768,512,true],"large":["https:\/\/www.tun.com\/home\/wp-content\/uploads\/2026\/10\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizer-1024x683.png",1024,683,true],"1536x1536":["https:\/\/www.tun.com\/home\/wp-content\/uploads\/2026\/10\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizer.png",1536,1024,false],"2048x2048":["https:\/\/www.tun.com\/home\/wp-content\/uploads\/2026\/10\/dharma-ais-vocabx-cuts-inference-cost-77-by-fixing-tokenizer.png",1536,1024,false]},"uagb_author_info":{"display_name":"The University Network","author_link":"https:\/\/www.tun.com\/home\/author\/funky_junkie\/"},"uagb_comment_info":0,"uagb_excerpt":"Researchers at Dharma AI found that the tokenizer \u2014 not the model \u2014 is one of the biggest cost levers in specialized AI applications. Their VocabX method cut measured inference cost by 77% in a production essay-grading system without changing the underlying model at all.","_links":{"self":[{"href":"https:\/\/www.tun.com\/home\/wp-json\/wp\/v2\/posts\/39861","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.tun.com\/home\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.tun.com\/home\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.tun.com\/home\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/www.tun.com\/home\/wp-json\/wp\/v2\/comments?post=39861"}],"version-history":[{"count":4,"href":"https:\/\/www.tun.com\/home\/wp-json\/wp\/v2\/posts\/39861\/revisions"}],"predecessor-version":[{"id":39894,"href":"https:\/\/www.tun.com\/home\/wp-json\/wp\/v2\/posts\/39861\/revisions\/39894"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.tun.com\/home\/wp-json\/wp\/v2\/media\/39860"}],"wp:attachment":[{"href":"https:\/\/www.tun.com\/home\/wp-json\/wp\/v2\/media?parent=39861"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.tun.com\/home\/wp-json\/wp\/v2\/categories?post=39861"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.tun.com\/home\/wp-json\/wp\/v2\/tags?post=39861"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}