{"slug": "ts-zip", "name": "ts_zip", "category": "other", "category_name": "Other software", "description": "", "recommenders": 4, "commenters": 9, "warners": 0, "score": 5.041, "rising": 1.35, "first": "2026-02-01", "latest": "2026-09-22", "why": "For something close to the state of the art see e.g. Fabrice Bellard (of course) ts_zip", "why_is_quote": true, "has_verdict": false, "sentiment": {"recommend": 4, "neutral": 4, "compare": 1, "warn": 0}, "spark": [0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 2, 0], "listed": true, "loved": [], "caveat": "", "short_quote": "For something close to the state of the art see e.g. Fabrice Bellard (of course) ts_zip", "kind": "", "icon": false, "domain": "bellard.org", "github_repo": null, "official_url": "https://bellard.org/ts_zip/", "pricing": "", "aliases": ["ts_zip"], "verdict": null, "praise": [{"label": "", "summary": "", "quotes": [{"id": 796910624, "item_id": 49806914, "url": "https://news.ycombinator.com/item?id=49806914", "by": "adamgordonbell", "date": "2026-09-22", "stance": "recommend", "quote": "Once you understand the connection, things like ts_zip and hutter prize make a lot more sense.", "excerpt": "Compressing something and next token prediction are very related. Once you understand the connection, things like ts_zip and hutter prize make a lot more sense. https://bellard.org/ts_zip/ https://corecursive.com/the-hutter-prize/", "context": "", "use_case": "", "switched_from": "", "story": "Can gzip be a language model?"}, {"id": 796767904, "item_id": 49797994, "url": "https://news.ycombinator.com/item?id=49797994", "by": "magicalhippo", "date": "2026-09-22", "stance": "recommend", "quote": "still beats xz by a solid margin", "excerpt": "> if LLMs are used as compressors, how well is that expected to work Quite well. This project[1], by Fabrice Bellard of ffmpeg fame, is quite old in AI years and uses an ancient LLM, but still beats xz by a solid margin. [1]: https://bellard.org/ts_zip/", "context": "", "use_case": "LLM-based compression", "switched_from": "", "story": "Can gzip be a language model?"}, {"id": 779990032, "item_id": 48749377, "url": "https://news.ycombinator.com/item?id=48749377", "by": "soulofmischief", "date": "2026-07-01", "stance": "recommend", "quote": "ts_zip compresses text at record efficiency", "excerpt": "I just expanded more in a reply to my own comment if you're interested! To technically expand a bit on AI: Any regularity in an environment which an embedded system can detect but fails to exploit represents an amount of excess free energy in the organism, distributed over itself, its group, its species, etc. depending on what types of systems and scales you choose to model. There are parallels in information theory:", "context": "", "use_case": "Text compression", "switched_from": "", "story": "Newly discovered spider builds spring loaded snare to catch ants"}, {"id": 777773424, "item_id": 48610839, "url": "https://news.ycombinator.com/item?id=48610839", "by": "sltkr", "date": "2026-06-20", "stance": "recommend", "quote": "For something close to the state of the art see e.g. Fabrice Bellard (of course) ts_zip", "excerpt": "compression = prediction + entropy coding was already an insight from Claude Shannon in the 1950s Since LLM are inherently token predictors, that makes using them for losless compression almost trivial. For something close to the state of the art see e.g. Fabrice Bellard (of course) ts_zip: https://bellard.org/ts_zip/ I think some of the confusion comes from the fact that there is a pretty big difference between the ", "context": "", "use_case": "lossless compression", "switched_from": "", "story": "Data Compression Explained"}]}], "caveats": [], "who": [], "tags": [], "use_cases": ["text compression", "llm-based compression", "lossless compression", "llm text compression"], "compared": [{"slug": "nncp", "name": "NNCP", "co": 2, "switched_from": 0, "switched_to": 0}, {"slug": "zstandard", "name": "Zstandard", "co": 2, "switched_from": 0, "switched_to": 0}], "makers": [], "threads": [{"url": "https://news.ycombinator.com/item?id=49797323", "title": "Can gzip be a language model?", "mentions": 2}, {"url": "https://news.ycombinator.com/item?id=49263497", "title": "Compression Is Prediction", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48711052", "title": "Newly discovered spider builds spring loaded snare to catch ants", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48562662", "title": "Data Compression Explained", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48550779", "title": "John Carmack on Fabrice Bellard", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=47565831", "title": "Show HN: I made a \"programming language\" looking for feedback", "mentions": 1}], "sentiment_by_month": {"2025-11": {}, "2025-12": {}, "2026-01": {}, "2026-02": {"compare": 1, "neutral": 1}, "2026-03": {"neutral": 1}, "2026-04": {}, "2026-05": {}, "2026-06": {"neutral": 1, "recommend": 1}, "2026-07": {"recommend": 1}, "2026-08": {"neutral": 1}, "2026-09": {"recommend": 2}, "2026-10": {}}, "quotes": [{"id": 796910624, "item_id": 49806914, "url": "https://news.ycombinator.com/item?id=49806914", "by": "adamgordonbell", "date": "2026-09-22", "stance": "recommend", "quote": "Once you understand the connection, things like ts_zip and hutter prize make a lot more sense.", "excerpt": "Compressing something and next token prediction are very related. Once you understand the connection, things like ts_zip and hutter prize make a lot more sense. https://bellard.org/ts_zip/ https://corecursive.com/the-hutter-prize/", "context": "", "use_case": "", "switched_from": "", "story": "Can gzip be a language model?"}, {"id": 796767904, "item_id": 49797994, "url": "https://news.ycombinator.com/item?id=49797994", "by": "magicalhippo", "date": "2026-09-22", "stance": "recommend", "quote": "still beats xz by a solid margin", "excerpt": "> if LLMs are used as compressors, how well is that expected to work Quite well. This project[1], by Fabrice Bellard of ffmpeg fame, is quite old in AI years and uses an ancient LLM, but still beats xz by a solid margin. [1]: https://bellard.org/ts_zip/", "context": "", "use_case": "LLM-based compression", "switched_from": "", "story": "Can gzip be a language model?"}, {"id": 788289072, "item_id": 49268067, "url": "https://news.ycombinator.com/item?id=49268067", "by": "vasco", "date": "2026-08-11", "stance": "neutral", "quote": "Fabrice Bellard published this in 2023", "excerpt": "Fabrice Bellard published this in 2023: https://bellard.org/ts_zip/ > The ts_zip utility can compress (and hopefully decompress) text files using a Large Language Model. The compression ratio is much higher than with other compression tools. It's not only an old idea it's been totally done already.", "context": "", "use_case": "", "switched_from": "", "story": "Compression Is Prediction"}, {"id": 779990032, "item_id": 48749377, "url": "https://news.ycombinator.com/item?id=48749377", "by": "soulofmischief", "date": "2026-07-01", "stance": "recommend", "quote": "ts_zip compresses text at record efficiency", "excerpt": "I just expanded more in a reply to my own comment if you're interested! To technically expand a bit on AI: Any regularity in an environment which an embedded system can detect but fails to exploit represents an amount of excess free energy in the organism, distributed over itself, its group, its species, etc. depending on what types of systems and scales you choose to model. There are parallels in information theory:", "context": "", "use_case": "Text compression", "switched_from": "", "story": "Newly discovered spider builds spring loaded snare to catch ants"}, {"id": 777773424, "item_id": 48610839, "url": "https://news.ycombinator.com/item?id=48610839", "by": "sltkr", "date": "2026-06-20", "stance": "recommend", "quote": "For something close to the state of the art see e.g. Fabrice Bellard (of course) ts_zip", "excerpt": "compression = prediction + entropy coding was already an insight from Claude Shannon in the 1950s Since LLM are inherently token predictors, that makes using them for losless compression almost trivial. For something close to the state of the art see e.g. Fabrice Bellard (of course) ts_zip: https://bellard.org/ts_zip/ I think some of the confusion comes from the fact that there is a pretty big difference between the ", "context": "", "use_case": "lossless compression", "switched_from": "", "story": "Data Compression Explained"}, {"id": 776867792, "item_id": 48554237, "url": "https://news.ycombinator.com/item?id=48554237", "by": "santiagobasulto", "date": "2026-06-16", "stance": "neutral", "quote": "Bellard has a very interesting project that is `ts_zip`, a compression algorithm powered by LLMs.", "excerpt": "Bellard has a very interesting project that is `ts_zip`, a compression algorithm powered by LLMs. It's just an \"experiment\" and should never be used in production, but very smart. The description on his website is amusing: \"The ts_zip utility can compress (and hopefully decompress) text files using a Large Language Model\" https://bellard.org/ts_zip/", "context": "", "use_case": "", "switched_from": "", "story": "John Carmack on Fabrice Bellard"}, {"id": 761103456, "item_id": 47568966, "url": "https://news.ycombinator.com/item?id=47568966", "by": "nextaccountic", "date": "2026-03-29", "stance": "neutral", "quote": "It beats stuff like .xz etc but of course is much slower.", "excerpt": "> The tech described in my comment's parent is speculative, and technically it's not even inference, once it's perfectly reproducible. This is not true. Fabrice Bellard's ts_zip [0] and ts_sms [1] uses a LLM to compress text. It beats stuff like .xz etc but of course is much slower. Now.. if it were non-deterministic, you would have trouble decompressing exactly into what it compressed. So, it uses a deterministic LL", "context": "", "use_case": "Text compression", "switched_from": "", "story": "Show HN: I made a \"programming language\" looking for feedback"}, {"id": 759610720, "item_id": 47475670, "url": "https://news.ycombinator.com/item?id=47475670", "by": "nextaccountic", "date": "2026-03-22", "stance": "neutral", "quote": "fabrice bellard's ts_zip", "excerpt": "The model reliably outputs nothing when prompted to embody the void. Anyway at a later date they concede that it's not 100% deterministic, because > Temperature 0 non-determinism. While all confirmatory results were 30/30, known floating-point non-determinism exists at temperature 0 in both APIs. One control concept (thunder) showed 1/30 void on GPT, demonstrating marginal non-determinism. Actually FP non-determinism", "context": "", "use_case": "", "switched_from": "", "story": "Cross-Model Void Convergence: GPT-5.2 and Claude Opus 4.6 Deterministic Silence"}, {"id": 751722352, "item_id": 46982647, "url": "https://news.ycombinator.com/item?id=46982647", "by": "dchest", "date": "2026-02-11", "stance": "neutral", "quote": "https://bellard.org/ts_zip/", "excerpt": "https://bellard.org/ts_zip/", "context": "", "use_case": "LLM text compression", "switched_from": "", "story": "Text classification with Python 3.14's ZSTD module"}, {"id": 749645696, "item_id": 46852856, "url": "https://news.ycombinator.com/item?id=46852856", "by": "boulos", "date": "2026-02-01", "stance": "compare", "quote": "Bellard's tz_zip (https://bellard.org/ts_zip/) or nncp compress really well they are extremely slow", "excerpt": "This isn't strictly correct: you probably mean wrt compressed size. Compression is a tradeoff between size reduction and compression and decompression speed. So while things like Bellard's tz_zip (https://bellard.org/ts_zip/) or nncp compress really well they are extremely slow compared to say zstd or the much faster compression scheme in the article. It's a totally different class of codec.", "context": "", "use_case": "", "switched_from": "", "story": "Efficient String Compression for Modern Database Systems"}]}