Hugging Face introduces ALTK Evolve to enhance Large Language model efficiency. Hugging Face and IBM Research unveil ALTK Evolve, a new method designed to significantly reduce token consumption in large language models while preserving performance, offering substantial benefits for developers and businesses. Published August 11, 2026 Hugging Face and IBM Research unveil ALTK Evolve for LLM efficiency. Large Language Models (LLMs) continue to advance rapidly, but their computational demands, particularly concerning token usage, remain a significant challenge. Addressing this, Hugging Face, in collaboration with IBM Research, has introduced ALTK Evolve. This innovative technique aims to streamline LLM operations by drastically reducing the number of tokens required for inference, promising greater efficiency without compromising model efficacy. The challenge of token efficiency in LLMs. The operational cost and speed of Large Language Models are heavily influenced by the number of tokens processed. Each token represents a piece of text or data, and every interaction with an LLM, from input prompts to generated responses, incurs a token cost. For developers and businesses deploying LLMs at scale, optimizing this token usage is paramount for managing expenses and improving response times. Traditional methods often involve trade-offs between efficiency and the quality of the model's output. Introducing ALTK Evolve: A novel approach. ALTK Evolve, which stands for "Adaptive Token-Length Knowledge Evolve," represents a significant step forward in tackling token inefficiency. The core idea behind ALTK Evolve is to intelligently compress the input sequence by identifying and removing redundant or less critical tokens before they are fed into the LLM. This preprocessing step ensures that the model only receives the most pertinent information, thereby reducing the computational load and associated costs. Unlike simpler truncation methods that might arbitrarily cut off parts of the input, ALTK Evolve employs a more sophisticated mechanism. It analyzes the input text, much like a human might skim for keywords, to prioritize and retain essential information. This selective reduction allows for a much shorter input sequence to be passed to the LLM, maintaining context and critical details crucial for accurate responses. How ALTK Evolve operates. The methodology behind ALTK Evolve involves a multi-stage process. Initially, the system processes the input text to understand its semantic structure and identify key information. This often involves leveraging smaller, specialized models or heuristics to determine token importance. Once the critical tokens are identified, the less relevant ones are effectively "evolved out" or removed, forming a condensed version of the original input. This condensed input is then fed into the target LLM for inference. During development, ALTK Evolve was rigorously tested across various benchmarks and tasks. The results indicated that models utilizing ALTK Evolve could achieve comparable performance to their full-input counterparts, but with a substantially lower token count. This demonstrates the technique's ability to maintain high-quality outputs while significantly cutting down on resource consumption. Implications for developers and businesses. The introduction of ALTK Evolve has profound implications for anyone working with or deploying LLMs. For developers, it means the ability to build more cost-effective and faster applications. Reducing token usage directly translates to lower API costs when interacting with commercial LLMs and faster inference times, which can enhance user experience in real-time applications. Businesses can benefit from increased throughput and reduced operational expenditures, making the deployment of sophisticated AI solutions more economically viable. Furthermore, by making LLMs more efficient, ALTK Evolve could enable the use of larger, more capable models in scenarios where computational constraints previously made them impractical. This allows for broader application of advanced AI, from complex data analysis to enhanced customer service chatbots. Why it matters. ALTK Evolve represents a critical advancement in the ongoing effort to make Large Language Models more practical and accessible. By addressing the fundamental challenge of token efficiency, Hugging Face and IBM Research are paving the way for more economical, faster, and sustainable AI deployments. This innovation has the potential to democratize access to powerful AI capabilities, allowing a wider range of developers and organizations to leverage state-of-the-art LLMs effectively and efficiently, ultimately fostering further innovation in the AI landscape.

97.6% on 300-year-old books. Then came the blank pages. August 11, 2026 · 00:12 UTC · News Tl;dr. On August 10, Hugging Face and EleutherAI published FineBooks, starting with a leaderboard that scores 14 open-weight OCR models on all 2,165 pages of an expert-transcribed corpus of 18th and 19th century natural history books. The top models read old print at roughly 97.6% character accuracy, and the whole run costs between $0.34 and $1.94 per 1,000 pages on rented GPUs. The interesting failure is not archaic type. It is that several models, handed a plate or a blank leaf, write paragraphs that were never there. The benchmark nobody had. Measuring OCR needs pages whose correct transcription is already known, character for character. Almost nobody has that for historical books. FineBooks found a set: between 2011 and 2012, the EU's IMPACT digitisation programme and BHL-Europe hand-corrected six volumes from the Biodiversity Heritage Library, 2,165 pages in English, French, German and Latin, down to roughly one error in two thousand characters. Both projects are long dead and their websites are gone. The transcriptions survived because they were released under CC-BY and preserved in a GitHub repository by IMPACT's technical manager, Clemens Neudecker. Somebody in 2012 picked an open license, which turns out to be the least glamorous form of time travel available to Ai Bacon: none of the models on this board existed then, and most of their architectures had not been invented. The rebuilt ground truth ships as finebooks/bhl-impact-gt, with every scan matched to its transcription, and the scorer ships as bhl-ocr-eval. One methodology note deserves to be copied by every other leaderboard: they ran every model themselves under a pinned container image, pinned model revision, pinned script commit and a recorded job id. Scores from hosted inference routers are refused outright, because a router will not tell you which provider served the request, at what quantization, or under what serving config, and character error rate moves with all three. The board. The headline metric is character error rate on the reading lane, which folds case, the long s and ligatures into their modern equivalents. dots.ocr and dots.mocr land at 2.35% and 2.37%, which the project's own results file calls statistically identical. OvisOCR2 takes third at 3.05%, PaddleOCR-VL-1.6 fourth at 3.92%. Now read that chart again from the bottom. Tesseract 5, a non-neural engine whose lineage predates the transformer by decades, comes in at 6.42% and lands within a point and a half of Qwen3.5-9B, the largest model on the board. It also never once fell into a repetition loop, which is more than any neural row can say. Cost splits the top differently than accuracy does. dots.mocr costs $1.94 per 1,000 pages on Hugging Face Jobs; PaddleOCR-VL-1.6 costs $0.34 for 96.1% accuracy. That is 5.7x the price for 1.5 points of characters, which is a trade you should make deliberately rather than by grabbing the top row. Params are measured, not quoted. The board sums every weight in each repo's safetensors index, vision towers included, instead of trusting the name. GLM-OCR, usually listed at 0.9B, is 1.33B. olmOCR-2-7B is 8.29B. Winning a size class is easier when you get to pick the number on the label. The blank-page problem. 428 of the 2,165 pages are sparse: engraved plates, blanks, near-empty leaves whose entire correct transcription might be one line of credits. The board scores them as their own stratum, and this is where the models stop resembling each other. Character error rate is a ratio: mistakes divided by the number of characters that should be on the page. When the denominator is tiny and the model invents text, the ratio blows past 1.0 and keeps going. A sparse CER of 26.93 means the model produced roughly 27 wrong characters for every real one. It is the OCR equivalent of an eyewitness giving a detailed description of a suspect at a crime scene where nobody was standing. That contrast is the finding. GLM-OCR reads ordinary content pages at 2.36% CER, better than everything except the dots pair, and then produces 26.93 on sparse pages. LightOnOCR-2 posts 15.79, olmOCR-2 posts 13.64, Qwen3.5-9B posts 19.98. Meanwhile dots.ocr sits at 1.79 and DeepSeek-OCR-2 at 2.09. If you are building a document pipeline, this is the column to weight, because real collections are full of plates, blanks, dividers and ads. An aggregate CER that averages over them will quietly hide a model that hallucinates whenever the page runs out of words. Why this matters if you never touch a library. Public domain books are one of the largest pools of long-form text that anyone can legally train on, and the text quality is currently set by whatever OCR tool a library happened to run once, at scan time, years ago. FineBooks cites the Talkie project's measurement of what that costs: a language model trained on OCR-derived text learned at 30% of the efficiency of the same model trained on human transcriptions of the same books. That reframes OCR from a digitisation chore into a training-data multiplier. When EleutherAI and collaborators shipped Common Pile, the largest openly licensed corpus to date, it carried around 300,000 public domain books whose text came from those earlier pipelines. Re-running them is, in the project's words, one of the highest-leverage improvements available to open training data. The next phase is already stated. The Biodiversity Heritage Library holds more than 300,000 digitized items and more than 64 million pages; roughly 200,000 of those items carry public domain rights statements, and FineBooks intends to re-run OCR across them with a leading model and publish the result as its first dataset. Do the arithmetic on their own cost figures and 64 million pages runs about $22,000 of GPU time at $0.34 per thousand, or about $124,000 at $1.94. The public domain slice is smaller than that, which puts a full re-processing somewhere between a grant line item and a rounding error in a frontier training budget. What the board does not tell you. The authors are unusually direct about scope, so take the caveats at full strength: * Four languages, one typeface family. English, French, German and Latin, all antiqua. The board says nothing about German Fraktur, non-Latin scripts, non-European languages or handwriting. * Books only. Single-column running text with consistent layout. Newspapers, magazines and archival material have different failure modes that are not measured here. * The confidence intervals are wide on purpose. The bootstrap resamples the six volumes rather than individual pages, so it is honest about book-to-book variation. Where intervals overlap, the models are tied, and the top two overlap. * Loop rate is a selection effect. Pages where a model repeats itself into its token cap are excluded from that row's other numbers, so heavy loopers look better than they are. PaddleOCR-VL-1.6 loops on 6.47% of pages; the top row is scored on 96.1% of the corpus. The verdict splits by use case. Good enough for LLM training corpora: yes, comfortably better than the legacy text sitting in open corpora today. Good enough for libraries replacing legacy OCR: it depends, because legacy pipelines emit ALTO XML with word-level coordinates and these models emit Markdown or plain text, sometimes with region boxes but never word positions. Good enough for scholarly transcription: not quite, since the models silently modernise the long s and ligatures that a diplomatic transcription exists to preserve. SmolDocling at 19.65% CER is a reminder that the size floor is real, too. Key takeaways. * FineBooks, a Hugging Face and EleutherAI collaboration, scored 14 open-weight OCR models on all 2,165 pages of expert-transcribed 18th and 19th century books, with the ground truth, harness and per-run provenance published alongside. * dots.ocr (2.35% CER) and dots.mocr (2.37%) lead and are statistically tied; the largest model on the board, Qwen3.5-9B at 9.65B parameters, finishes eighth at 5.07%. * On the 428 sparse or blank pages, several models invent text wholesale: GLM-OCR scores 2.36% on content pages and 26.93 on sparse ones, with Qwen3.5-9B at 19.98 and olmOCR-2 at 13.64. * Cost runs $0.34 to $1.94 per 1,000 pages on rented GPUs, which makes re-OCR of an entire collection a budgeting question rather than a research question. * The motivation is training data: a model trained on OCR-derived text learned at 30% of the efficiency of one trained on human transcriptions, and Common Pile carries around 300,000 books produced by older pipelines. * Scope is narrow by design: antiqua typefaces, four languages, single-column books, no Fraktur, no handwriting, no newspapers. AI OCR Open Weights Hugging Face EleutherAI Benchmarks Datasets Training Data AI Bacon Weekly Liked this? Get smarter about AI, weekly. The model launches that matter, the local-LLM tricks worth stealing, and the tools actually shipping. Hand-curated into one email a week, cut down to signal. No hype, ever. Frontier models Local LLMs Builder tools Zero hype Free and curated. One tap to unsubscribe, always. Error: Domain verification failed (missing Origin header).