{"slug": "gemma-4-26b-a4b", "name": "Gemma 4 26B A4B", "category": "ai", "category_name": "AI models and assistants", "description": "Google's open-weight mixture-of-experts language model, 26B total and about 4B active parameters.", "recommenders": 9, "commenters": 11, "warners": 1, "score": 7.562, "rising": -1.62, "first": "2026-05-17", "latest": "2026-10-16", "why": "HN commenters like it as a fast local model on Macs and consumer GPUs for writing help, language study and simpler coding tasks.", "why_is_quote": false, "has_verdict": true, "sentiment": {"recommend": 10, "neutral": 0, "compare": 1, "warn": 1}, "spark": [0, 0, 0, 0, 0, 0, 0, 5, 3, 1, 1, 0], "listed": true, "loved": ["Simpler coding help", "Everyday writing tasks"], "caveat": "Unreliable coding", "short_quote": "For simpler programming related (e.g. text2sql) as well as synthetic data generation, current best for me is llama.cpp + Gemma-4-26B-A4B", "icon": true, "domain": null, "github_repo": null, "official_url": "https://ai.google.dev/gemma", "pricing": "free", "aliases": ["Gemma 4 26B A4B", "Gemma 4 26B", "Gemma 4 26B-A4B", "Gemma-4-26B-A4B"], "verdict": [{"text": "HN commenters like it as a fast local model on Macs and consumer GPUs", "cites": [{"id": 790516336, "item_id": 49407271, "url": "https://news.ycombinator.com/item?id=49407271", "by": "LeBit", "date": "2026-08-23", "stance": "recommend", "quote": "Gemma 4 26B A4B gives me steady 20 tps.", "excerpt": "It\u2019s not because you didn\u2019t find use cases for local LLMs that there are none. I use local LLMs on my Mac Mini M4 Pro with 48G to review text messages tone, act as a text correction tool, act as a code review tool, to do code agent work, generate code snippets, etc Gemma 4 26B A4B gives me steady 20 tps.", "context": "Mac Mini M4 Pro 48GB user", "use_case": "local LLM tasks on Mac", "switched_from": "", "story": "Why your local LLM feels dumber than it is"}, {"id": 781849312, "item_id": 48865582, "url": "https://news.ycombinator.com/item?id=48865582", "by": "swe_dima", "date": "2026-07-10", "stance": "recommend", "quote": "I have switched to gemma-4-26b-a4b.\n\nAt least in benchmarks, it scores higher and is faster.", "excerpt": "the writing is on the wall for it, I have switched to gemma-4-26b-a4b. At least in benchmarks, it scores higher and is faster.", "context": "", "use_case": "", "switched_from": "Gemini 2.5 Flash", "story": "Don't discontinue Gemini 2.5 Flash"}, {"id": 776713696, "item_id": 48544606, "url": "https://news.ycombinator.com/item?id=48544606", "by": "argee", "date": "2026-06-15", "stance": "recommend", "quote": "it's fast and good enough that it could handle being a \"co-pilot\" on small to medium context tasks", "excerpt": "I use Gemma 4 26B A4B on my Macbook (M4 Pro, 48 GB RAM) to study Rust (and ask other myriad questions). I don't trust it to do a good job in an IDE/harness to one-shot anything but the most trivial of changes. Still, it's fast and good enough that it could handle being a \"co-pilot\" on small to medium context tasks where you've got your hands on the wheel and your eyes on the road \u2014 and are driving under the speed lim", "context": "uses a Macbook M4 Pro with 48 GB RAM", "use_case": "studying Rust and Q&A locally", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}, {"id": 776825859, "item_id": 48551616, "url": "https://news.ycombinator.com/item?id=48551616", "by": "ljosifov", "date": "2026-06-16", "stance": "recommend", "quote": "For simpler programming related (e.g. text2sql) as well as synthetic data generation, current best for me is llama.cpp + Gemma-4-26B-A4B", "excerpt": "Not replaced but supplemented. For off-line coding current setup is pi + ds4-server + DeepSeek-V4-Flash REAP25 (on M2 Max 96gb). For simpler programming related (e.g. text2sql) as well as synthetic data generation, current best for me is llama.cpp + Gemma-4-26B-A4B (on gpu 7900xtx 24gb; sometimes nemotron-cascade-2-30b-a3b for 1M context). That and (dabbling now) auto-research uses lots of tokens. Used to get paused ", "context": "", "use_case": "text2sql, synthetic data", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}]}, {"text": " for writing help, language study and simpler coding tasks.", "cites": [{"id": 797379842, "item_id": 49836240, "url": "https://news.ycombinator.com/item?id=49836240", "by": "LeBit", "date": "2026-09-24", "stance": "recommend", "quote": "I find them quite useful!", "excerpt": "I use the following models: - Qwen3.5 9B - Qwen3.6 35B A3B - Qwen3.8 27B - Gemma 4 26B A4B - Gemma 31B - Muse Glimmer 30B I have 48G. The MacMini is used solely for inference. llama.cpp + llama-swap. People are saying local models are crap and serves no purposes. I use them to help me spell check, write emails, write text messages, write JIRA tickets, write PR comments, etc. I also use them in coding agents to comple", "context": "Mac Mini with 48G", "use_case": "local inference", "switched_from": "", "story": "Best LLM for every budget, updated daily"}, {"id": 785593649, "item_id": 49099603, "url": "https://news.ycombinator.com/item?id=49099603", "by": "ktosobcy", "date": "2026-07-29", "stance": "recommend", "quote": "`gemma-4-26b-a4b` is suprisingly capable", "excerpt": "Recently I started playing with LM Studio and local models and found out that `gemma-4-26b-a4b` is suprisingly capable. I don't need elaborate akin to \"create complete app to do X\" or \"refactor the whole codebase of bazzilion of LOC\" but rather \"how to go about doing thing X\" or for language study (explaining nuances of phrasal verbs or subtelties of vocabulary in other languagues) and darn -- the results are rather ", "context": "", "use_case": "coding questions, language study", "switched_from": "", "story": "Self-hosting Kimi K3: 20% more hardware cost, 20% better task resolution"}, {"id": 776825859, "item_id": 48551616, "url": "https://news.ycombinator.com/item?id=48551616", "by": "ljosifov", "date": "2026-06-16", "stance": "recommend", "quote": "For simpler programming related (e.g. text2sql) as well as synthetic data generation, current best for me is llama.cpp + Gemma-4-26B-A4B", "excerpt": "Not replaced but supplemented. For off-line coding current setup is pi + ds4-server + DeepSeek-V4-Flash REAP25 (on M2 Max 96gb). For simpler programming related (e.g. text2sql) as well as synthetic data generation, current best for me is llama.cpp + Gemma-4-26B-A4B (on gpu 7900xtx 24gb; sometimes nemotron-cascade-2-30b-a3b for 1M context). That and (dabbling now) auto-research uses lots of tokens. Used to get paused ", "context": "", "use_case": "text2sql, synthetic data", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}, {"id": 776714241, "item_id": 48544640, "url": "https://news.ycombinator.com/item?id=48544640", "by": "jmichaelson", "date": "2026-06-15", "stance": "recommend", "quote": "It's a very good model when the harness keeps it on rails.", "excerpt": "I am working on exactly this issue right now. My approach is that a highly optimized harness (pi.dev) with the right backing knowledgebase (a custom, self-updating wiki with lots of QC layers) can get close to most of my usage patterns for my Claude Max 20x subscription. I use Gemma 4 26B QAT served by a custom fork of llama.cpp, with 4-8 slots of 256k context at Q8. It's a very good model when the harness keeps it o", "context": "scientific programming", "use_case": "local coding model", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}]}, {"text": " The most common caveat is that it is weak at unsupervised coding", "cites": [{"id": 801896960, "item_id": 50118560, "url": "https://news.ycombinator.com/item?id=50118560", "by": "blinkingled", "date": "2026-10-16", "stance": "warn", "quote": "I used Gemma4 26B before and it wasn't good in my testing.", "excerpt": "Yes 27B - not so much of a surprise but I was hoping for a usable coding model that's local and fast - this is fast but maybe not usable. I used Gemma4 26B before and it wasn't good in my testing.", "context": "", "use_case": "local coding model", "switched_from": "", "story": "LFM2.5-24B-A2B: Built for fast on-device agents"}, {"id": 776713696, "item_id": 48544606, "url": "https://news.ycombinator.com/item?id=48544606", "by": "argee", "date": "2026-06-15", "stance": "recommend", "quote": "it's fast and good enough that it could handle being a \"co-pilot\" on small to medium context tasks", "excerpt": "I use Gemma 4 26B A4B on my Macbook (M4 Pro, 48 GB RAM) to study Rust (and ask other myriad questions). I don't trust it to do a good job in an IDE/harness to one-shot anything but the most trivial of changes. Still, it's fast and good enough that it could handle being a \"co-pilot\" on small to medium context tasks where you've got your hands on the wheel and your eyes on the road \u2014 and are driving under the speed lim", "context": "uses a Macbook M4 Pro with 48 GB RAM", "use_case": "studying Rust and Q&A locally", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}]}, {"text": " and needs a human or a tight harness to keep it on track.", "cites": [{"id": 776713696, "item_id": 48544606, "url": "https://news.ycombinator.com/item?id=48544606", "by": "argee", "date": "2026-06-15", "stance": "recommend", "quote": "it's fast and good enough that it could handle being a \"co-pilot\" on small to medium context tasks", "excerpt": "I use Gemma 4 26B A4B on my Macbook (M4 Pro, 48 GB RAM) to study Rust (and ask other myriad questions). I don't trust it to do a good job in an IDE/harness to one-shot anything but the most trivial of changes. Still, it's fast and good enough that it could handle being a \"co-pilot\" on small to medium context tasks where you've got your hands on the wheel and your eyes on the road \u2014 and are driving under the speed lim", "context": "uses a Macbook M4 Pro with 48 GB RAM", "use_case": "studying Rust and Q&A locally", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}, {"id": 776714241, "item_id": 48544640, "url": "https://news.ycombinator.com/item?id=48544640", "by": "jmichaelson", "date": "2026-06-15", "stance": "recommend", "quote": "It's a very good model when the harness keeps it on rails.", "excerpt": "I am working on exactly this issue right now. My approach is that a highly optimized harness (pi.dev) with the right backing knowledgebase (a custom, self-updating wiki with lots of QC layers) can get close to most of my usage patterns for my Claude Max 20x subscription. I use Gemma 4 26B QAT served by a custom fork of llama.cpp, with 4-8 slots of 256k context at Q8. It's a very good model when the harness keeps it o", "context": "scientific programming", "use_case": "local coding model", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}]}], "praise": [{"label": "Simpler coding help", "summary": "Used for code review, text2sql, snippets and co-pilot style work on small to medium tasks.", "quotes": [{"id": 776825859, "item_id": 48551616, "url": "https://news.ycombinator.com/item?id=48551616", "by": "ljosifov", "date": "2026-06-16", "stance": "recommend", "quote": "For simpler programming related (e.g. text2sql) as well as synthetic data generation, current best for me is llama.cpp + Gemma-4-26B-A4B", "excerpt": "Not replaced but supplemented. For off-line coding current setup is pi + ds4-server + DeepSeek-V4-Flash REAP25 (on M2 Max 96gb). For simpler programming related (e.g. text2sql) as well as synthetic data generation, current best for me is llama.cpp + Gemma-4-26B-A4B (on gpu 7900xtx 24gb; sometimes nemotron-cascade-2-30b-a3b for 1M context). That and (dabbling now) auto-research uses lots of tokens. Used to get paused ", "context": "", "use_case": "text2sql, synthetic data", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}, {"id": 776713696, "item_id": 48544606, "url": "https://news.ycombinator.com/item?id=48544606", "by": "argee", "date": "2026-06-15", "stance": "recommend", "quote": "it's fast and good enough that it could handle being a \"co-pilot\" on small to medium context tasks", "excerpt": "I use Gemma 4 26B A4B on my Macbook (M4 Pro, 48 GB RAM) to study Rust (and ask other myriad questions). I don't trust it to do a good job in an IDE/harness to one-shot anything but the most trivial of changes. Still, it's fast and good enough that it could handle being a \"co-pilot\" on small to medium context tasks where you've got your hands on the wheel and your eyes on the road \u2014 and are driving under the speed lim", "context": "uses a Macbook M4 Pro with 48 GB RAM", "use_case": "studying Rust and Q&A locally", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}, {"id": 776714241, "item_id": 48544640, "url": "https://news.ycombinator.com/item?id=48544640", "by": "jmichaelson", "date": "2026-06-15", "stance": "recommend", "quote": "It's a very good model when the harness keeps it on rails.", "excerpt": "I am working on exactly this issue right now. My approach is that a highly optimized harness (pi.dev) with the right backing knowledgebase (a custom, self-updating wiki with lots of QC layers) can get close to most of my usage patterns for my Claude Max 20x subscription. I use Gemma 4 26B QAT served by a custom fork of llama.cpp, with 4-8 slots of 256k context at Q8. It's a very good model when the harness keeps it o", "context": "scientific programming", "use_case": "local coding model", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}]}, {"label": "Everyday writing tasks", "summary": "Handles spell checking, emails, tickets, message tone review and text correction.", "quotes": [{"id": 797379842, "item_id": 49836240, "url": "https://news.ycombinator.com/item?id=49836240", "by": "LeBit", "date": "2026-09-24", "stance": "recommend", "quote": "I find them quite useful!", "excerpt": "I use the following models: - Qwen3.5 9B - Qwen3.6 35B A3B - Qwen3.8 27B - Gemma 4 26B A4B - Gemma 31B - Muse Glimmer 30B I have 48G. The MacMini is used solely for inference. llama.cpp + llama-swap. People are saying local models are crap and serves no purposes. I use them to help me spell check, write emails, write text messages, write JIRA tickets, write PR comments, etc. I also use them in coding agents to comple", "context": "Mac Mini with 48G", "use_case": "local inference", "switched_from": "", "story": "Best LLM for every budget, updated daily"}, {"id": 790516336, "item_id": 49407271, "url": "https://news.ycombinator.com/item?id=49407271", "by": "LeBit", "date": "2026-08-23", "stance": "recommend", "quote": "Gemma 4 26B A4B gives me steady 20 tps.", "excerpt": "It\u2019s not because you didn\u2019t find use cases for local LLMs that there are none. I use local LLMs on my Mac Mini M4 Pro with 48G to review text messages tone, act as a text correction tool, act as a code review tool, to do code agent work, generate code snippets, etc Gemma 4 26B A4B gives me steady 20 tps.", "context": "Mac Mini M4 Pro 48GB user", "use_case": "local LLM tasks on Mac", "switched_from": "", "story": "Why your local LLM feels dumber than it is"}]}, {"label": "Speed", "summary": "Commenters describe it as fast, with one reporting a steady 20 tokens per second on an M4 Pro.", "quotes": [{"id": 790516336, "item_id": 49407271, "url": "https://news.ycombinator.com/item?id=49407271", "by": "LeBit", "date": "2026-08-23", "stance": "recommend", "quote": "Gemma 4 26B A4B gives me steady 20 tps.", "excerpt": "It\u2019s not because you didn\u2019t find use cases for local LLMs that there are none. I use local LLMs on my Mac Mini M4 Pro with 48G to review text messages tone, act as a text correction tool, act as a code review tool, to do code agent work, generate code snippets, etc Gemma 4 26B A4B gives me steady 20 tps.", "context": "Mac Mini M4 Pro 48GB user", "use_case": "local LLM tasks on Mac", "switched_from": "", "story": "Why your local LLM feels dumber than it is"}, {"id": 781849312, "item_id": 48865582, "url": "https://news.ycombinator.com/item?id=48865582", "by": "swe_dima", "date": "2026-07-10", "stance": "recommend", "quote": "I have switched to gemma-4-26b-a4b.\n\nAt least in benchmarks, it scores higher and is faster.", "excerpt": "the writing is on the wall for it, I have switched to gemma-4-26b-a4b. At least in benchmarks, it scores higher and is faster.", "context": "", "use_case": "", "switched_from": "Gemini 2.5 Flash", "story": "Don't discontinue Gemini 2.5 Flash"}, {"id": 776713696, "item_id": 48544606, "url": "https://news.ycombinator.com/item?id=48544606", "by": "argee", "date": "2026-06-15", "stance": "recommend", "quote": "it's fast and good enough that it could handle being a \"co-pilot\" on small to medium context tasks", "excerpt": "I use Gemma 4 26B A4B on my Macbook (M4 Pro, 48 GB RAM) to study Rust (and ask other myriad questions). I don't trust it to do a good job in an IDE/harness to one-shot anything but the most trivial of changes. Still, it's fast and good enough that it could handle being a \"co-pilot\" on small to medium context tasks where you've got your hands on the wheel and your eyes on the road \u2014 and are driving under the speed lim", "context": "uses a Macbook M4 Pro with 48 GB RAM", "use_case": "studying Rust and Q&A locally", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}]}, {"label": "Learning and questions", "summary": "Used for language study and learning Rust, with results described as rather good.", "quotes": [{"id": 785593649, "item_id": 49099603, "url": "https://news.ycombinator.com/item?id=49099603", "by": "ktosobcy", "date": "2026-07-29", "stance": "recommend", "quote": "`gemma-4-26b-a4b` is suprisingly capable", "excerpt": "Recently I started playing with LM Studio and local models and found out that `gemma-4-26b-a4b` is suprisingly capable. I don't need elaborate akin to \"create complete app to do X\" or \"refactor the whole codebase of bazzilion of LOC\" but rather \"how to go about doing thing X\" or for language study (explaining nuances of phrasal verbs or subtelties of vocabulary in other languagues) and darn -- the results are rather ", "context": "", "use_case": "coding questions, language study", "switched_from": "", "story": "Self-hosting Kimi K3: 20% more hardware cost, 20% better task resolution"}, {"id": 776713696, "item_id": 48544606, "url": "https://news.ycombinator.com/item?id=48544606", "by": "argee", "date": "2026-06-15", "stance": "recommend", "quote": "it's fast and good enough that it could handle being a \"co-pilot\" on small to medium context tasks", "excerpt": "I use Gemma 4 26B A4B on my Macbook (M4 Pro, 48 GB RAM) to study Rust (and ask other myriad questions). I don't trust it to do a good job in an IDE/harness to one-shot anything but the most trivial of changes. Still, it's fast and good enough that it could handle being a \"co-pilot\" on small to medium context tasks where you've got your hands on the wheel and your eyes on the road \u2014 and are driving under the speed lim", "context": "uses a Macbook M4 Pro with 48 GB RAM", "use_case": "studying Rust and Q&A locally", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}]}, {"label": "Runs on local hardware", "summary": "Offers a good balance for small local setups and is preferred by some over a similar Qwen model.", "quotes": [{"id": 777381649, "item_id": 48586353, "url": "https://news.ycombinator.com/item?id=48586353", "by": "jeidoban", "date": "2026-06-18", "stance": "recommend", "quote": "I find that Gemma 4 26b a4b gives the best balance here", "excerpt": "Been working on an open source project for a while called Ironsmith. The gist is you can create highly specific macOS apps for yourself with just a prompt, and one of my main goals from the beginning was to get it to work with low end models like the Apple foundation and the Gemma series. It uses a custom agentic loop tailor made to work with small models with limited context. This means you can create very simple ap", "context": "", "use_case": "generating Mac apps locally", "switched_from": "", "story": "Show HN: Ironsmith \u2013 Generate personal native Mac apps with local LLMs"}, {"id": 774861536, "item_id": 48428846, "url": "https://news.ycombinator.com/item?id=48428846", "by": "ColonelPhantom", "date": "2026-06-06", "stance": "recommend", "quote": "though I use Gemma 4 26B-A4B more than Qwen3.6 35B-A3B", "excerpt": "Deepseek V4 Flash still has 13B active params though? That is about half as many as Qwen3.6-27B (and much more than Qwen3.6-35B-A3B). Given that RAM (even on a base M4 or 'regular' Intel/AMD system) is like an order of magnitude faster than an SSD, even Qwen 27B running from RAM will be much faster than any Deepseek V4 model with SSD offloading. And the MoE will be much faster still. Qwen 27B is also small enough to ", "context": "runs models on a Ryzen 7 7840U laptop with 64 GB RAM", "use_case": "local inference on laptop", "switched_from": "", "story": "Nvidia is proposing a beast of a CPU system for Windows PCs"}]}, {"label": "Simple architecture", "summary": "One engine builder chose it over Qwen for its simpler architecture and kernels.", "quotes": [{"id": 785717472, "item_id": 49107342, "url": "https://news.ycombinator.com/item?id=49107342", "by": "gitpusher42", "date": "2026-07-30", "stance": "recommend", "quote": "I switched to Gemma because of its simpler architecture and kernels", "excerpt": "Yeah, the same ideas should work for qwen. You can try porting this engine to use Owen. Owen 3.6-35b-a3b was my initial idea, but I switched to Gemma because of its simpler architecture and kernels", "context": "built an open-source inference engine", "use_case": "local inference engine target", "switched_from": "Qwen 3.6-35B-A3B", "story": "Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac"}]}], "caveats": [{"label": "Unreliable coding", "summary": "Commenters report it is not good for coding on its own and should not be trusted to one-shot changes.", "quotes": [{"id": 801896960, "item_id": 50118560, "url": "https://news.ycombinator.com/item?id=50118560", "by": "blinkingled", "date": "2026-10-16", "stance": "warn", "quote": "I used Gemma4 26B before and it wasn't good in my testing.", "excerpt": "Yes 27B - not so much of a surprise but I was hoping for a usable coding model that's local and fast - this is fast but maybe not usable. I used Gemma4 26B before and it wasn't good in my testing.", "context": "", "use_case": "local coding model", "switched_from": "", "story": "LFM2.5-24B-A2B: Built for fast on-device agents"}, {"id": 776713696, "item_id": 48544606, "url": "https://news.ycombinator.com/item?id=48544606", "by": "argee", "date": "2026-06-15", "stance": "recommend", "quote": "it's fast and good enough that it could handle being a \"co-pilot\" on small to medium context tasks", "excerpt": "I use Gemma 4 26B A4B on my Macbook (M4 Pro, 48 GB RAM) to study Rust (and ask other myriad questions). I don't trust it to do a good job in an IDE/harness to one-shot anything but the most trivial of changes. Still, it's fast and good enough that it could handle being a \"co-pilot\" on small to medium context tasks where you've got your hands on the wheel and your eyes on the road \u2014 and are driving under the speed lim", "context": "uses a Macbook M4 Pro with 48 GB RAM", "use_case": "studying Rust and Q&A locally", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}]}, {"label": "Needs a harness", "summary": "One user says it is very good only when a harness keeps it on rails.", "quotes": [{"id": 776714241, "item_id": 48544640, "url": "https://news.ycombinator.com/item?id=48544640", "by": "jmichaelson", "date": "2026-06-15", "stance": "recommend", "quote": "It's a very good model when the harness keeps it on rails.", "excerpt": "I am working on exactly this issue right now. My approach is that a highly optimized harness (pi.dev) with the right backing knowledgebase (a custom, self-updating wiki with lots of QC layers) can get close to most of my usage patterns for my Claude Max 20x subscription. I use Gemma 4 26B QAT served by a custom fork of llama.cpp, with 4-8 slots of 256k context at Q8. It's a very good model when the harness keeps it o", "context": "scientific programming", "use_case": "local coding model", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}]}, {"label": "Stronger alternatives", "summary": "One commenter recommends Qwen3.6 models first and suggests Gemma only if it suits your tasks better.", "quotes": [{"id": 770705555, "item_id": 48169097, "url": "https://news.ycombinator.com/item?id=48169097", "by": "sleepyeldrazi", "date": "2026-05-17", "stance": "compare", "quote": "(or gemma 4 26B if gemma models perform better for your tasks)", "excerpt": "If you want a good dense model, use qwen3.6 27B instead, speed will be up, and if you don't take my word for it being smarter, take openrouter's prices of it against the bigger, slower and less memory-efficient gemma do the talking. If you want a faster model, go for qwen3.6 35B (or gemma 4 26B if gemma models perform better for your tasks). There is a reason why people (myself included) haven't shut up about those t", "context": "", "use_case": "", "switched_from": "", "story": "Apple Silicon costs more than OpenRouter"}]}], "who": [{"by": "LeBit", "context": "Mac Mini with 48G", "url": "https://news.ycombinator.com/item?id=49836240", "date": "2026-09-24"}, {"by": "gitpusher42", "context": "built an open-source inference engine", "url": "https://news.ycombinator.com/item?id=49107342", "date": "2026-07-30"}, {"by": "jmichaelson", "context": "scientific programming", "url": "https://news.ycombinator.com/item?id=48544640", "date": "2026-06-15"}, {"by": "argee", "context": "uses a Macbook M4 Pro with 48 GB RAM", "url": "https://news.ycombinator.com/item?id=48544606", "date": "2026-06-15"}, {"by": "ColonelPhantom", "context": "runs models on a Ryzen 7 7840U laptop with 64 GB RAM", "url": "https://news.ycombinator.com/item?id=48428846", "date": "2026-06-06"}], "tags": ["for local inference", "for Mac users", "for simple coding tasks", "for language study"], "use_cases": ["local coding model", "local inference", "local llm tasks on mac", "local inference engine target", "coding questions, language study", "generating mac apps locally"], "compared": [{"slug": "qwen-3-6", "name": "Qwen 3.6", "co": 4, "switched_from": 1, "switched_to": 0}, {"slug": "llama-cpp", "name": "llama.cpp", "co": 4, "switched_from": 0, "switched_to": 0}, {"slug": "pi", "name": "Pi", "co": 3, "switched_from": 0, "switched_to": 0}, {"slug": "qwen-3-5", "name": "Qwen 3.5", "co": 1, "switched_from": 0, "switched_to": 1}, {"slug": "gemini-2-5-flash", "name": "Gemini 2.5 Flash", "co": 1, "switched_from": 1, "switched_to": 0}, {"slug": "ollama", "name": "Ollama", "co": 2, "switched_from": 0, "switched_to": 0}, {"slug": "deepseek-v4-flash", "name": "DeepSeek V4 Flash", "co": 2, "switched_from": 0, "switched_to": 0}, {"slug": "claude", "name": "Claude", "co": 2, "switched_from": 0, "switched_to": 0}], "makers": [], "threads": [{"url": "https://news.ycombinator.com/item?id=48542100", "title": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?", "mentions": 3}, {"url": "https://news.ycombinator.com/item?id=50097369", "title": "LFM2.5-24B-A2B: Built for fast on-device agents", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49830866", "title": "Best LLM for every budget, updated daily", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49402232", "title": "Why your local LLM feels dumber than it is", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49098510", "title": "Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49098130", "title": "Self-hosting Kimi K3: 20% more hardware cost, 20% better task resolution", "mentions": 1}], "sentiment_by_month": {"2025-11": {}, "2025-12": {}, "2026-01": {}, "2026-02": {}, "2026-03": {}, "2026-04": {}, "2026-05": {"compare": 1}, "2026-06": {"recommend": 5}, "2026-07": {"recommend": 3}, "2026-08": {"recommend": 1}, "2026-09": {"recommend": 1}, "2026-10": {"warn": 1}}, "quotes": [{"id": 801896960, "item_id": 50118560, "url": "https://news.ycombinator.com/item?id=50118560", "by": "blinkingled", "date": "2026-10-16", "stance": "warn", "quote": "I used Gemma4 26B before and it wasn't good in my testing.", "excerpt": "Yes 27B - not so much of a surprise but I was hoping for a usable coding model that's local and fast - this is fast but maybe not usable. I used Gemma4 26B before and it wasn't good in my testing.", "context": "", "use_case": "local coding model", "switched_from": "", "story": "LFM2.5-24B-A2B: Built for fast on-device agents"}, {"id": 797379842, "item_id": 49836240, "url": "https://news.ycombinator.com/item?id=49836240", "by": "LeBit", "date": "2026-09-24", "stance": "recommend", "quote": "I find them quite useful!", "excerpt": "I use the following models: - Qwen3.5 9B - Qwen3.6 35B A3B - Qwen3.8 27B - Gemma 4 26B A4B - Gemma 31B - Muse Glimmer 30B I have 48G. The MacMini is used solely for inference. llama.cpp + llama-swap. People are saying local models are crap and serves no purposes. I use them to help me spell check, write emails, write text messages, write JIRA tickets, write PR comments, etc. I also use them in coding agents to comple", "context": "Mac Mini with 48G", "use_case": "local inference", "switched_from": "", "story": "Best LLM for every budget, updated daily"}, {"id": 790516336, "item_id": 49407271, "url": "https://news.ycombinator.com/item?id=49407271", "by": "LeBit", "date": "2026-08-23", "stance": "recommend", "quote": "Gemma 4 26B A4B gives me steady 20 tps.", "excerpt": "It\u2019s not because you didn\u2019t find use cases for local LLMs that there are none. I use local LLMs on my Mac Mini M4 Pro with 48G to review text messages tone, act as a text correction tool, act as a code review tool, to do code agent work, generate code snippets, etc Gemma 4 26B A4B gives me steady 20 tps.", "context": "Mac Mini M4 Pro 48GB user", "use_case": "local LLM tasks on Mac", "switched_from": "", "story": "Why your local LLM feels dumber than it is"}, {"id": 785717472, "item_id": 49107342, "url": "https://news.ycombinator.com/item?id=49107342", "by": "gitpusher42", "date": "2026-07-30", "stance": "recommend", "quote": "I switched to Gemma because of its simpler architecture and kernels", "excerpt": "Yeah, the same ideas should work for qwen. You can try porting this engine to use Owen. Owen 3.6-35b-a3b was my initial idea, but I switched to Gemma because of its simpler architecture and kernels", "context": "built an open-source inference engine", "use_case": "local inference engine target", "switched_from": "Qwen 3.6-35B-A3B", "story": "Show HN: Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac"}, {"id": 785593649, "item_id": 49099603, "url": "https://news.ycombinator.com/item?id=49099603", "by": "ktosobcy", "date": "2026-07-29", "stance": "recommend", "quote": "`gemma-4-26b-a4b` is suprisingly capable", "excerpt": "Recently I started playing with LM Studio and local models and found out that `gemma-4-26b-a4b` is suprisingly capable. I don't need elaborate akin to \"create complete app to do X\" or \"refactor the whole codebase of bazzilion of LOC\" but rather \"how to go about doing thing X\" or for language study (explaining nuances of phrasal verbs or subtelties of vocabulary in other languagues) and darn -- the results are rather ", "context": "", "use_case": "coding questions, language study", "switched_from": "", "story": "Self-hosting Kimi K3: 20% more hardware cost, 20% better task resolution"}, {"id": 781849312, "item_id": 48865582, "url": "https://news.ycombinator.com/item?id=48865582", "by": "swe_dima", "date": "2026-07-10", "stance": "recommend", "quote": "I have switched to gemma-4-26b-a4b.\n\nAt least in benchmarks, it scores higher and is faster.", "excerpt": "the writing is on the wall for it, I have switched to gemma-4-26b-a4b. At least in benchmarks, it scores higher and is faster.", "context": "", "use_case": "", "switched_from": "Gemini 2.5 Flash", "story": "Don't discontinue Gemini 2.5 Flash"}, {"id": 777381649, "item_id": 48586353, "url": "https://news.ycombinator.com/item?id=48586353", "by": "jeidoban", "date": "2026-06-18", "stance": "recommend", "quote": "I find that Gemma 4 26b a4b gives the best balance here", "excerpt": "Been working on an open source project for a while called Ironsmith. The gist is you can create highly specific macOS apps for yourself with just a prompt, and one of my main goals from the beginning was to get it to work with low end models like the Apple foundation and the Gemma series. It uses a custom agentic loop tailor made to work with small models with limited context. This means you can create very simple ap", "context": "", "use_case": "generating Mac apps locally", "switched_from": "", "story": "Show HN: Ironsmith \u2013 Generate personal native Mac apps with local LLMs"}, {"id": 776825859, "item_id": 48551616, "url": "https://news.ycombinator.com/item?id=48551616", "by": "ljosifov", "date": "2026-06-16", "stance": "recommend", "quote": "For simpler programming related (e.g. text2sql) as well as synthetic data generation, current best for me is llama.cpp + Gemma-4-26B-A4B", "excerpt": "Not replaced but supplemented. For off-line coding current setup is pi + ds4-server + DeepSeek-V4-Flash REAP25 (on M2 Max 96gb). For simpler programming related (e.g. text2sql) as well as synthetic data generation, current best for me is llama.cpp + Gemma-4-26B-A4B (on gpu 7900xtx 24gb; sometimes nemotron-cascade-2-30b-a3b for 1M context). That and (dabbling now) auto-research uses lots of tokens. Used to get paused ", "context": "", "use_case": "text2sql, synthetic data", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}, {"id": 776714241, "item_id": 48544640, "url": "https://news.ycombinator.com/item?id=48544640", "by": "jmichaelson", "date": "2026-06-15", "stance": "recommend", "quote": "It's a very good model when the harness keeps it on rails.", "excerpt": "I am working on exactly this issue right now. My approach is that a highly optimized harness (pi.dev) with the right backing knowledgebase (a custom, self-updating wiki with lots of QC layers) can get close to most of my usage patterns for my Claude Max 20x subscription. I use Gemma 4 26B QAT served by a custom fork of llama.cpp, with 4-8 slots of 256k context at Q8. It's a very good model when the harness keeps it o", "context": "scientific programming", "use_case": "local coding model", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}, {"id": 776713696, "item_id": 48544606, "url": "https://news.ycombinator.com/item?id=48544606", "by": "argee", "date": "2026-06-15", "stance": "recommend", "quote": "it's fast and good enough that it could handle being a \"co-pilot\" on small to medium context tasks", "excerpt": "I use Gemma 4 26B A4B on my Macbook (M4 Pro, 48 GB RAM) to study Rust (and ask other myriad questions). I don't trust it to do a good job in an IDE/harness to one-shot anything but the most trivial of changes. Still, it's fast and good enough that it could handle being a \"co-pilot\" on small to medium context tasks where you've got your hands on the wheel and your eyes on the road \u2014 and are driving under the speed lim", "context": "uses a Macbook M4 Pro with 48 GB RAM", "use_case": "studying Rust and Q&A locally", "switched_from": "", "story": "Ask HN: Has anyone replaced Claude/GPT with a local model for daily coding?"}, {"id": 774861536, "item_id": 48428846, "url": "https://news.ycombinator.com/item?id=48428846", "by": "ColonelPhantom", "date": "2026-06-06", "stance": "recommend", "quote": "though I use Gemma 4 26B-A4B more than Qwen3.6 35B-A3B", "excerpt": "Deepseek V4 Flash still has 13B active params though? That is about half as many as Qwen3.6-27B (and much more than Qwen3.6-35B-A3B). Given that RAM (even on a base M4 or 'regular' Intel/AMD system) is like an order of magnitude faster than an SSD, even Qwen 27B running from RAM will be much faster than any Deepseek V4 model with SSD offloading. And the MoE will be much faster still. Qwen 27B is also small enough to ", "context": "runs models on a Ryzen 7 7840U laptop with 64 GB RAM", "use_case": "local inference on laptop", "switched_from": "", "story": "Nvidia is proposing a beast of a CPU system for Windows PCs"}, {"id": 770705555, "item_id": 48169097, "url": "https://news.ycombinator.com/item?id=48169097", "by": "sleepyeldrazi", "date": "2026-05-17", "stance": "compare", "quote": "(or gemma 4 26B if gemma models perform better for your tasks)", "excerpt": "If you want a good dense model, use qwen3.6 27B instead, speed will be up, and if you don't take my word for it being smarter, take openrouter's prices of it against the bigger, slower and less memory-efficient gemma do the talking. If you want a faster model, go for qwen3.6 35B (or gemma 4 26B if gemma models perform better for your tasks). There is a reason why people (myself included) haven't shut up about those t", "context": "", "use_case": "", "switched_from": "", "story": "Apple Silicon costs more than OpenRouter"}]}