{"slug": "bitnet", "name": "BitNet", "category": "ai", "category_name": "AI models and assistants", "description": "", "recommenders": 3, "commenters": 9, "warners": 2, "score": 1.953, "rising": -0.98, "first": "2026-01-06", "latest": "2026-08-20", "why": "", "why_is_quote": false, "has_verdict": false, "sentiment": {"recommend": 2, "neutral": 4, "compare": 1, "warn": 2}, "spark": [0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0], "listed": true, "loved": [], "caveat": "", "short_quote": "", "kind": "", "icon": true, "domain": null, "github_repo": "github.com/microsoft/bitnet", "official_url": "https://github.com/microsoft/BitNet", "pricing": "", "aliases": ["BitNet"], "verdict": null, "praise": [{"label": "", "summary": "", "quotes": [{"id": 784338560, "item_id": 49021160, "url": "https://news.ycombinator.com/item?id=49021160", "by": "ForHackernews", "date": "2026-07-23", "stance": "recommend", "quote": "https://github.com/microsoft/BitNet", "excerpt": "No, because the LLMs will keep getting more efficient and capable. Distillation and quantization will mean firms spending trillions on giant data centres are left holding the bag. I suspect Apple ends up laughing all the way to the bank. https://github.com/microsoft/BitNet", "context": "", "use_case": "", "switched_from": "", "story": "Alphabet's cash burn raises alarm for Big Tech as AI spending climbs"}, {"id": 760495233, "item_id": 47530952, "url": "https://news.ycombinator.com/item?id=47530952", "by": "naasking", "date": "2026-03-26", "stance": "recommend", "quote": "Hopefully Microsoft keeps pushing BitNet too", "excerpt": "Newer quantization approaches are even better, 4-bits gets you no meaningful loss relative to FP16: https://github.com/z-lab/paroquant Hopefully Microsoft keeps pushing BitNet too, so only \"1.58\" bits are needed. I think fractional representations are only relevant for training at this point, and bf16 is sufficient, no need for fp4 and such.", "context": "", "use_case": "", "switched_from": "", "story": "Quantization from the Ground Up"}, {"id": 757366320, "item_id": 47335395, "url": "https://news.ycombinator.com/item?id=47335395", "by": "Arcuru", "date": "2026-03-11", "stance": "recommend", "quote": "I thought it showed a lot of promise", "excerpt": "It's good to see this getting some continued development. I looked into it last year[1] and I thought it showed a lot of promise so I've been very disappointed that I never saw a newer model. [1] - https://jackson.dev/post/dont-sleep-on-bitnet/", "context": "", "use_case": "", "switched_from": "", "story": "Microsoft BitNet: 100B Param 1-Bit model for local CPUs"}]}], "caveats": [{"label": "", "summary": "", "quotes": [{"id": 757502192, "item_id": 47343887, "url": "https://news.ycombinator.com/item?id=47343887", "by": "kristopolous", "date": "2026-03-11", "stance": "warn", "quote": "BitNet really doesn't perform that well on evals.", "excerpt": "I don't see the news here ... there's https://huggingface.co/collections/microsoft/bitnet which is last updated 12/2025 ... am I just paying more attention here or is there something actually new about this? Also as far as I know, this is more of a research curiosity - BitNet really doesn't perform that well on evals. I think Qwen3.5 2B is the best you can get in the ~1GB class.", "context": "", "use_case": "", "switched_from": "", "story": "Microsoft BitNet: 100B Param 1-Bit model for local CPUs"}, {"id": 757373984, "item_id": 47335874, "url": "https://news.ycombinator.com/item?id=47335874", "by": "webXL", "date": "2026-03-11", "stance": "warn", "quote": "It comes from (intentionally?) misleading docs", "excerpt": "It comes from (intentionally?) misleading docs: https://github.com/microsoft/BitNet/issues/391 (only suggesting that it's intentional because it's been there so long)", "context": "", "use_case": "1-bit LLM inference", "switched_from": "", "story": "Microsoft BitNet: 100B Param 1-Bit model for local CPUs"}]}], "who": [], "tags": [], "use_cases": ["ternary llm inference", "cpu inference for bitnet models", "1-bit model inference on cpu", "1-bit llm inference"], "compared": [{"slug": "llama-cpp", "name": "llama.cpp", "co": 2, "switched_from": 0, "switched_to": 0}, {"slug": "qwen", "name": "Qwen", "co": 2, "switched_from": 0, "switched_to": 0}], "makers": [], "threads": [{"url": "https://news.ycombinator.com/item?id=47334694", "title": "Microsoft BitNet: 100B Param 1-Bit model for local CPUs", "mentions": 5}, {"url": "https://news.ycombinator.com/item?id=49378441", "title": "AVX Bitwise ternary logic [VPTERNLOGD] instruction busted (2024)", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49258111", "title": "I wrote a C inference engine that runs BitNet 2B at 15\u201318 tok/s on a Pi 5", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49021006", "title": "Alphabet's cash burn raises alarm for Big Tech as AI spending climbs", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48085821", "title": "Local AI needs to be the norm", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=47519295", "title": "Quantization from the Ground Up", "mentions": 1}], "sentiment_by_month": {"2025-11": {}, "2025-12": {}, "2026-01": {"compare": 1}, "2026-02": {}, "2026-03": {"recommend": 1, "warn": 2, "neutral": 1}, "2026-04": {}, "2026-05": {"neutral": 1}, "2026-06": {}, "2026-07": {"recommend": 1}, "2026-08": {"neutral": 2}, "2026-09": {}, "2026-10": {}}, "quotes": [{"id": 790059888, "item_id": 49378743, "url": "https://news.ycombinator.com/item?id=49378743", "by": "peter_d_sherman", "date": "2026-08-20", "stance": "neutral", "quote": "Microsoft BitNet - Official inference framework for 1-bit LLMs", "excerpt": "This article (comparing the x86 AVX-512F instruction VPTERNLOGD to Amiga's Blitter) is not the end of the discussion for this instruction, it's actually the beginning... See, VPTERNLOGD is a ternary bitwise instruction. In theory, that means that it could (in theory, temporarily ignoring other factors like streaming, parallelism, memory access speed, etc.) replace the Matrix Multiply (aka \"MatMul\") component of LLM's", "context": "", "use_case": "ternary LLM inference", "switched_from": "", "story": "AVX Bitwise ternary logic [VPTERNLOGD] instruction busted (2024)"}, {"id": 788130193, "item_id": 49258137, "url": "https://news.ycombinator.com/item?id=49258137", "by": "geisten", "date": "2026-08-11", "stance": "neutral", "quote": "it runs Microsoft\u2019s BitNet 2B model fully offline", "excerpt": "Hi HN \u2014 I spent the last three months building an LLM inference engine from scratch in C. The result is *geistlib*: it runs Microsoft\u2019s BitNet 2B model fully offline on a 4 GB Raspberry Pi 5, packaged as a single ~1.2 GB executable containing both the engine and the model. Install and run: ``` curl -L -o geist-bitnet https://github.com/geisten/geistlib/releases/latest/download/geist-bitnet-linux-arm64 chmod +x geist-", "context": "", "use_case": "", "switched_from": "", "story": "I wrote a C inference engine that runs BitNet 2B at 15\u201318 tok/s on a Pi 5"}, {"id": 784338560, "item_id": 49021160, "url": "https://news.ycombinator.com/item?id=49021160", "by": "ForHackernews", "date": "2026-07-23", "stance": "recommend", "quote": "https://github.com/microsoft/BitNet", "excerpt": "No, because the LLMs will keep getting more efficient and capable. Distillation and quantization will mean firms spending trillions on giant data centres are left holding the bag. I suspect Apple ends up laughing all the way to the bank. https://github.com/microsoft/BitNet", "context": "", "use_case": "", "switched_from": "", "story": "Alphabet's cash burn raises alarm for Big Tech as AI spending climbs"}, {"id": 769470624, "item_id": 48091914, "url": "https://news.ycombinator.com/item?id=48091914", "by": "josteink", "date": "2026-05-11", "stance": "neutral", "quote": "Microsoft has researched 1-bit LLMs which can run much more efficiently, and on much cheaper hardware", "excerpt": "> You are greatly underestimating the current hardware requirements for productive local LLMs. Fixed that for you. Right now most models produced are based on floating point maths and probabilities, which is \"expensive\" to do math on. Microsoft has researched 1-bit LLMs which can run much more efficiently, and on much cheaper hardware[1]. If this research is reproducable and reusable outside their research models, th", "context": "", "use_case": "", "switched_from": "", "story": "Local AI needs to be the norm"}, {"id": 760495233, "item_id": 47530952, "url": "https://news.ycombinator.com/item?id=47530952", "by": "naasking", "date": "2026-03-26", "stance": "recommend", "quote": "Hopefully Microsoft keeps pushing BitNet too", "excerpt": "Newer quantization approaches are even better, 4-bits gets you no meaningful loss relative to FP16: https://github.com/z-lab/paroquant Hopefully Microsoft keeps pushing BitNet too, so only \"1.58\" bits are needed. I think fractional representations are only relevant for training at this point, and bf16 is sufficient, no need for fp4 and such.", "context": "", "use_case": "", "switched_from": "", "story": "Quantization from the Ground Up"}, {"id": 757514816, "item_id": 47344676, "url": "https://news.ycombinator.com/item?id=47344676", "by": "naasking", "date": "2026-03-11", "stance": "neutral", "quote": "they recently released a new inference engine for CPUs for bitnet models", "excerpt": "I don't think there is a new model, they recently released a new inference engine for CPUs for bitnet models: https://github.com/microsoft/BitNet/blob/main/src/README.md", "context": "", "use_case": "CPU inference for bitnet models", "switched_from": "", "story": "Microsoft BitNet: 100B Param 1-Bit model for local CPUs"}, {"id": 757502192, "item_id": 47343887, "url": "https://news.ycombinator.com/item?id=47343887", "by": "kristopolous", "date": "2026-03-11", "stance": "warn", "quote": "BitNet really doesn't perform that well on evals.", "excerpt": "I don't see the news here ... there's https://huggingface.co/collections/microsoft/bitnet which is last updated 12/2025 ... am I just paying more attention here or is there something actually new about this? Also as far as I know, this is more of a research curiosity - BitNet really doesn't perform that well on evals. I think Qwen3.5 2B is the best you can get in the ~1GB class.", "context": "", "use_case": "", "switched_from": "", "story": "Microsoft BitNet: 100B Param 1-Bit model for local CPUs"}, {"id": 757387248, "item_id": 47336703, "url": "https://news.ycombinator.com/item?id=47336703", "by": "naasking", "date": "2026-03-11", "stance": "neutral", "quote": "showing linear speedups with number of threads", "excerpt": "I think the README [1] for the new CPU feature is of more interest, showing linear speedups with number of threads. Up to 73 tokens/sec with 8 threads (64 toks/s for their recommended Q6 quant): https://github.com/microsoft/BitNet/blob/main/src/README.md", "context": "", "use_case": "1-bit model inference on CPU", "switched_from": "", "story": "Microsoft BitNet: 100B Param 1-Bit model for local CPUs"}, {"id": 757373984, "item_id": 47335874, "url": "https://news.ycombinator.com/item?id=47335874", "by": "webXL", "date": "2026-03-11", "stance": "warn", "quote": "It comes from (intentionally?) misleading docs", "excerpt": "It comes from (intentionally?) misleading docs: https://github.com/microsoft/BitNet/issues/391 (only suggesting that it's intentional because it's been there so long)", "context": "", "use_case": "1-bit LLM inference", "switched_from": "", "story": "Microsoft BitNet: 100B Param 1-Bit model for local CPUs"}, {"id": 757366320, "item_id": 47335395, "url": "https://news.ycombinator.com/item?id=47335395", "by": "Arcuru", "date": "2026-03-11", "stance": "recommend", "quote": "I thought it showed a lot of promise", "excerpt": "It's good to see this getting some continued development. I looked into it last year[1] and I thought it showed a lot of promise so I've been very disappointed that I never saw a newer model. [1] - https://jackson.dev/post/dont-sleep-on-bitnet/", "context": "", "use_case": "", "switched_from": "", "story": "Microsoft BitNet: 100B Param 1-Bit model for local CPUs"}, {"id": 744339730, "item_id": 46521233, "url": "https://news.ycombinator.com/item?id=46521233", "by": "LargoLasskhyfv", "date": "2026-01-06", "stance": "compare", "quote": "Or maybe even Microsofts Bitnet?", "excerpt": "Have you tried anything with https://codeberg.org/ikawrakow/illama https://github.com/ikawrakow/ik_llama.cpp and their 4Bit-quants? Or maybe even Microsofts Bitnet? https://github.com/microsoft/BitNet https://github.com/ikawrakow/ik_llama.cpp/pull/337 https://huggingface.co/microsoft/bitnet-b1.58-2B-4T-gguf ? That would be an interesting comparison for running local LLMs on such low-end/edge-devices. Or common office", "context": "", "use_case": "", "switched_from": "", "story": "A 30B Qwen Model Walks into a Raspberry Pi and Runs in Real Time"}]}