{"slug": "llama", "name": "Llama", "category": "ai", "category_name": "AI models and assistants", "description": "", "recommenders": 3, "commenters": 16, "warners": 4, "score": 1.825, "rising": 1.67, "first": "2025-11-01", "latest": "2026-10-20", "why": "I've actually seen better performance with older ones like Meta Llama 4 or Amazon's Nova", "why_is_quote": true, "has_verdict": false, "sentiment": {"recommend": 3, "neutral": 9, "compare": 0, "warn": 4}, "spark": [0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 1], "listed": true, "loved": [], "caveat": "", "short_quote": "I've actually seen better performance with older ones like Meta Llama 4 or Amazon's Nova", "icon": true, "domain": "simonwillison.net", "github_repo": null, "official_url": "https://www.llama.com", "pricing": "", "aliases": ["Llama", "Llama 4", "Llama 3", "Llama 3.3 70B", "Llama 3.1", "LLaMA"], "verdict": null, "praise": [{"label": "", "summary": "", "quotes": [{"id": 802691361, "item_id": 50168210, "url": "https://news.ycombinator.com/item?id=50168210", "by": "Alifatisk", "date": "2026-10-20", "stance": "recommend", "quote": "Did you forget what sparked the whole open-weight wave? It was that.", "excerpt": "> These people have arguably contributed nothing to society for over a decade [\u2026] they\u2019re being praised for being the example of best engineers. React? We can all criticize it, but it has revolutionized the web. Llama. Did you forget what sparked the whole open-weight wave? It was that. I dislike Meta, but I would not criticize the engineers competence. They have done impressive things, check out their blog if you ge", "context": "", "use_case": "", "switched_from": "", "story": "Facebook Pages require paid subscription to post more than two links per month"}, {"id": 786421169, "item_id": 49151323, "url": "https://news.ycombinator.com/item?id=49151323", "by": "andreidbr", "date": "2026-08-02", "stance": "recommend", "quote": "I've actually seen better performance with older ones like Meta Llama 4 or Amazon's Nova", "excerpt": "I've worked on a LLM-based visual validation framework for mobile layouts of different websites. The results are pretty flaky, even with Claude models. I've actually seen better performance with older ones like Meta Llama 4 or Amazon's Nova (the solution was running on AWS Bedrock). I've since refactored it to first do some deterministic checks using CSS, to determine visual glitches like overlaps. Only after do I us", "context": "", "use_case": "visual validation of mobile layouts", "switched_from": "", "story": "Boris Cherny on Trying to Get Claude Code to Rewrite the Claude App"}, {"id": 752558465, "item_id": 47034904, "url": "https://news.ycombinator.com/item?id=47034904", "by": "andsoitis", "date": "2026-02-16", "stance": "recommend", "quote": "You can run smaller models locally that are pretty good at code gen.", "excerpt": "You can run smaller models locally that are pretty good at code gen. Give it a try. To get started: https://simonwillison.net/2024/Nov/12/qwen25-coder/ or https://simonwillison.net/2024/Dec/9/llama-33-70b/", "context": "", "use_case": "local code generation", "switched_from": "", "story": "Ask HN: What happens after the AI bubble bursts?"}]}], "caveats": [{"label": "", "summary": "", "quotes": [{"id": 763180224, "item_id": 47698764, "url": "https://news.ycombinator.com/item?id=47698764", "by": "nl", "date": "2026-04-08", "stance": "warn", "quote": "Llama 4 was a bad model, unfortunately.", "excerpt": "> They were ~gpt4o, with the added benefit that you could run them on premise. No, they are bad models. They were benchmaxxed on LMAreana and a few other benchmarks but as soon as you try them yourself they fall to pieces. I have my own agentic benchmark[1] I use to compare models. Llama-4-scout-17b-16e scores 14/25, while llama-4-maverick-17b-128e scores 12/25. By comparison gemma-4-E4B-it-GGUF:Q4_K_M scores 15/25 (", "context": "runs own agentic benchmark", "use_case": "agentic benchmark comparison", "switched_from": "", "story": "Muse Spark \u2013 Meta Superintelligence Labs"}, {"id": 758280338, "item_id": 47392521, "url": "https://news.ycombinator.com/item?id=47392521", "by": "Lama9901", "date": "2026-03-15", "stance": "warn", "quote": "Meta confirmed Llama 4 benchmarks were fudged.", "excerpt": "Update \u2014 March 15, 2026. I owe this thread a more honest picture. When I posted this, the repo had 91 tests and 5 claims. Today: 14 claims across 7 domains (ML, materials science, digital twin, pharma, finance, IoT), 223 passing tests, 3 independent verification layers, and a Cross-Claim Cryptographic Chain that links physical constants (E = 70 GPa for aluminum) to simulation outputs \u2014 cryptographically, end-to-end. ", "context": "", "use_case": "", "switched_from": "", "story": null}, {"id": 741096464, "item_id": 46318529, "url": "https://news.ycombinator.com/item?id=46318529", "by": "eibrahim", "date": "2025-12-18", "stance": "warn", "quote": "I tried llama a few months back and the output was very disappointing.", "excerpt": "I tried llama a few months back and the output was very disappointing. I don't think it will handle book writing well or produce good quality. what do you mean tts for interviews? You mean use AI to interview grandpa and get his story and then use that to create \"his story\"?", "context": "", "use_case": "book writing", "switched_from": "", "story": "Ask HN: How would you monetize an AI book-writing app?"}, {"id": 735736308, "item_id": 45983519, "url": "https://news.ycombinator.com/item?id=45983519", "by": "mesadb", "date": "2025-11-19", "stance": "warn", "quote": "It looked nice but the latency was not acceptable for everyday use.", "excerpt": "Some implementation details, since getting this to work well was not trivial. My goal was \u201cpress hotkey, start talking, see text within ~1\u20132 seconds\u201d on an M2 MacBook Pro, and support multiple languages. First attempts (cloud) \u2013 I tried Hugging Face real-time transcription. It worked but latency was all over the place and costs would not scale. \u2013 I tried OpenAI real-time transcription. Latency was better, but when th", "context": "built a macOS dictation app", "use_case": "formatting transcripts into email", "switched_from": "", "story": "Show HN: Slave \u2013 local dictation and TTS for macOS (3k words free)"}]}], "who": [{"by": "ryptophan", "context": "Tech Lead at Georgian", "url": "https://news.ycombinator.com/item?id=49550977", "date": "2026-09-03"}, {"by": "Jignesh_1313", "context": "final-year engineer building AI projects", "url": "https://news.ycombinator.com/item?id=47690225", "date": "2026-04-08"}, {"by": "junyuren", "context": "author of Mappa", "url": "https://news.ycombinator.com/item?id=46889907", "date": "2026-02-04"}, {"by": "trsthales", "context": "maker", "url": "https://news.ycombinator.com/item?id=46667540", "date": "2026-01-18"}], "tags": [], "use_cases": ["rewriting benchmark queries", "local inference", "local llm use", "sentiment analysis", "visual validation of mobile layouts", "agentic benchmark comparison"], "compared": [{"slug": "gemma", "name": "Gemma", "co": 3, "switched_from": 0, "switched_to": 0}, {"slug": "qwen", "name": "Qwen", "co": 3, "switched_from": 0, "switched_to": 0}, {"slug": "postgresql", "name": "PostgreSQL", "co": 2, "switched_from": 0, "switched_to": 0}, {"slug": "claude", "name": "Claude", "co": 2, "switched_from": 0, "switched_to": 0}, {"slug": "gemini", "name": "Gemini", "co": 2, "switched_from": 0, "switched_to": 0}, {"slug": "llama-cpp", "name": "llama.cpp", "co": 2, "switched_from": 0, "switched_to": 0}, {"slug": "react", "name": "React", "co": 2, "switched_from": 0, "switched_to": 0}, {"slug": "ollama", "name": "Ollama", "co": 2, "switched_from": 0, "switched_to": 0}], "makers": [], "threads": [{"url": "https://news.ycombinator.com/item?id=50167536", "title": "Facebook Pages require paid subscription to post more than two links per month", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49550976", "title": "Your Queries Can Reorder the Retrieval Leaderboard", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49479951", "title": "Run Qwen3.8 27B locally: real numbers from my Mac Studio", "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=49200652", "title": "Qwen3.8 Max now ranked as the best overall model by agentic index", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49149800", "title": "Boris Cherny on Trying to Get Claude Code to Rewrite the Claude App", "mentions": 1}], "sentiment_by_month": {"2025-11": {"neutral": 1, "warn": 1}, "2025-12": {"warn": 1, "neutral": 1}, "2026-01": {"neutral": 1}, "2026-02": {"neutral": 1, "recommend": 1}, "2026-03": {"warn": 1}, "2026-04": {"neutral": 1, "warn": 1}, "2026-05": {}, "2026-06": {}, "2026-07": {}, "2026-08": {"recommend": 1, "neutral": 3}, "2026-09": {"neutral": 1}, "2026-10": {"recommend": 1}}, "quotes": [{"id": 802691361, "item_id": 50168210, "url": "https://news.ycombinator.com/item?id=50168210", "by": "Alifatisk", "date": "2026-10-20", "stance": "recommend", "quote": "Did you forget what sparked the whole open-weight wave? It was that.", "excerpt": "> These people have arguably contributed nothing to society for over a decade [\u2026] they\u2019re being praised for being the example of best engineers. React? We can all criticize it, but it has revolutionized the web. Llama. Did you forget what sparked the whole open-weight wave? It was that. I dislike Meta, but I would not criticize the engineers competence. They have done impressive things, check out their blog if you ge", "context": "", "use_case": "", "switched_from": "", "story": "Facebook Pages require paid subscription to post more than two links per month"}, {"id": 792815635, "item_id": 49550977, "url": "https://news.ycombinator.com/item?id=49550977", "by": "ryptophan", "date": "2026-09-03", "stance": "neutral", "quote": "gemini-2.5-pro, claude-opus-4.8, gpt-5.1, llama-3.3-70b", "excerpt": "Author here. I'm a Tech Lead at Georgian (a growth-stage VC). Short version: I used a few retrieval benchmark datasets, and experimented with a 21 embedding model leaderboard. I held the benchmark corpus and relevance labels fixed, and transformed the queries: paraphrase, terse, verbose. I kept the \u201cinformation need\u201d the same under each query transformation, but the leaderboard order changed more than I expected. I r", "context": "Tech Lead at Georgian", "use_case": "rewriting benchmark queries", "switched_from": "", "story": "Your Queries Can Reorder the Retrieval Leaderboard"}, {"id": 791773010, "item_id": 49485813, "url": "https://news.ycombinator.com/item?id=49485813", "by": "wolvoleo", "date": "2026-08-29", "stance": "neutral", "quote": "llama 3.1 8b q8_0 does about 70tk/s", "excerpt": "I use Llama.cpp and OpenAI-style API of course (who doesn't, unless you use Ollama perhaps?) Ollama sucks, they dropped ROCm support for my card out of the blue, and the Vulkan replacement wasn't ready. The developer just shrugged and didn't care about the people affected so I moved to llama-server. Also, here's some other reasons to avoid ollama: https://sleepingrobots.com/dreams/stop-using-ollama/ And really, ollam", "context": "", "use_case": "local inference", "switched_from": "", "story": "Run Qwen3.8 27B locally: real numbers from my Mac Studio"}, {"id": 790526657, "item_id": 49407916, "url": "https://news.ycombinator.com/item?id=49407916", "by": "synthrakx", "date": "2026-08-23", "stance": "neutral", "quote": "I have been using open source LLMs locally for the past 5 months like Qwen, Llama, DeepSeek and others", "excerpt": "I have been using open source LLMs locally for the past 5 months like Qwen, Llama, DeepSeek and others, and I have also noticed that current local models feel significantly dumber than closed source commercial models like ChatGPT, Claude, and Gemini. One main reason I think is the amount, variety, and quality of original authentic data on which they are being trained on, and also the training method plays a significa", "context": "", "use_case": "local LLM use", "switched_from": "", "story": "Why your local LLM feels dumber than it is"}, {"id": 787267220, "item_id": 49204201, "url": "https://news.ycombinator.com/item?id=49204201", "by": "vunderba", "date": "2026-08-06", "stance": "neutral", "quote": "Some small 8b models (like llama3.1) for sentiment analysis on text.", "excerpt": "Oh I do. Headless 128GB RAM machine serving llama.cpp with a number of local models that I use on a daily basis. \u2022 Qwen3-VL picks up new images in a NAS, auto captions and adds the text descriptions as a hidden EXIF layer into the image, which is used for fast search and organization in conjunction with a Qdrant vector database. \u2022 Gemma3:27b is used for personal translation work (mostly English and Chinese). \u2022 Some s", "context": "", "use_case": "sentiment analysis", "switched_from": "", "story": "Qwen3.8 Max now ranked as the best overall model by agentic index"}, {"id": 786421169, "item_id": 49151323, "url": "https://news.ycombinator.com/item?id=49151323", "by": "andreidbr", "date": "2026-08-02", "stance": "recommend", "quote": "I've actually seen better performance with older ones like Meta Llama 4 or Amazon's Nova", "excerpt": "I've worked on a LLM-based visual validation framework for mobile layouts of different websites. The results are pretty flaky, even with Claude models. I've actually seen better performance with older ones like Meta Llama 4 or Amazon's Nova (the solution was running on AWS Bedrock). I've since refactored it to first do some deterministic checks using CSS, to determine visual glitches like overlaps. Only after do I us", "context": "", "use_case": "visual validation of mobile layouts", "switched_from": "", "story": "Boris Cherny on Trying to Get Claude Code to Rewrite the Claude App"}, {"id": 763180224, "item_id": 47698764, "url": "https://news.ycombinator.com/item?id=47698764", "by": "nl", "date": "2026-04-08", "stance": "warn", "quote": "Llama 4 was a bad model, unfortunately.", "excerpt": "> They were ~gpt4o, with the added benefit that you could run them on premise. No, they are bad models. They were benchmaxxed on LMAreana and a few other benchmarks but as soon as you try them yourself they fall to pieces. I have my own agentic benchmark[1] I use to compare models. Llama-4-scout-17b-16e scores 14/25, while llama-4-maverick-17b-128e scores 12/25. By comparison gemma-4-E4B-it-GGUF:Q4_K_M scores 15/25 (", "context": "runs own agentic benchmark", "use_case": "agentic benchmark comparison", "switched_from": "", "story": "Muse Spark \u2013 Meta Superintelligence Labs"}, {"id": 763043606, "item_id": 47690225, "url": "https://news.ycombinator.com/item?id=47690225", "by": "Jignesh_1313", "date": "2026-04-08", "stance": "neutral", "quote": "a LangGraph + Groq Llama 3.3 70B agent that reads GitHub PRs and issues", "excerpt": "The multi-agent + RAG combo is what I'm living in right now. At AmberFlux EdgeAI I'm building SHEELA \u2014 an intelligence layer that crawls, scores, and runs controlled self-improvement loops on top of a FastAPI/PostgreSQL infrastructure backbone I architected from scratch. LangGraph for the agent orchestration, async pipelines throughout. But the project that feels most relevant to what you're describing is something I", "context": "final-year engineer building AI projects", "use_case": "PR and issue analysis agent", "switched_from": "", "story": "Ask HN: Who is hiring? (April 2026)"}, {"id": 758280338, "item_id": 47392521, "url": "https://news.ycombinator.com/item?id=47392521", "by": "Lama9901", "date": "2026-03-15", "stance": "warn", "quote": "Meta confirmed Llama 4 benchmarks were fudged.", "excerpt": "Update \u2014 March 15, 2026. I owe this thread a more honest picture. When I posted this, the repo had 91 tests and 5 claims. Today: 14 claims across 7 domains (ML, materials science, digital twin, pharma, finance, IoT), 223 passing tests, 3 independent verification layers, and a Cross-Claim Cryptographic Chain that links physical constants (E = 70 GPa for aluminum) to simulation outputs \u2014 cryptographically, end-to-end. ", "context": "", "use_case": "", "switched_from": "", "story": null}, {"id": 752558465, "item_id": 47034904, "url": "https://news.ycombinator.com/item?id=47034904", "by": "andsoitis", "date": "2026-02-16", "stance": "recommend", "quote": "You can run smaller models locally that are pretty good at code gen.", "excerpt": "You can run smaller models locally that are pretty good at code gen. Give it a try. To get started: https://simonwillison.net/2024/Nov/12/qwen25-coder/ or https://simonwillison.net/2024/Dec/9/llama-33-70b/", "context": "", "use_case": "local code generation", "switched_from": "", "story": "Ask HN: What happens after the AI bubble bursts?"}, {"id": 750238515, "item_id": 46889907, "url": "https://news.ycombinator.com/item?id=46889907", "by": "junyuren", "date": "2026-02-04", "stance": "neutral", "quote": "We tested with Qwen and LLaMA base models.", "excerpt": "Author here. Happy to answer questions. The problem: when you have multiple LLM agents working together and something fails, which agent is responsible? Traditional RL gives you one reward at the end, so all agents share the blame equally. Our approach: an external LLM (we used Gemini) watches each agent's actions and tool outputs, then assigns per-action scores. When agent 3 crashes because agent 1 forgot to save a ", "context": "author of Mappa", "use_case": "base models for training", "switched_from": "", "story": "Mappa \u2013 Fine-tune ANY multi-agent LLM systems end-to-end with AI coaches"}, {"id": 746680641, "item_id": 46667540, "url": "https://news.ycombinator.com/item?id=46667540", "by": "trsthales", "date": "2026-01-18", "stance": "neutral", "quote": "I use Llama 3 to translate natural language rules into \"JSON Logic\"", "excerpt": "Hello HN, maker here. I built StackForge because I was tired of the \"Day 1\" problem: setting up Auth, Stripe, Prisma, and Docker configurations for every new SaaS idea. Most low-code tools lock you into their ecosystem. I wanted an \"accelerator\" that outputs standard, clean code that I can eject and host anywhere. The Stack: - Frontend: React 19 + Vite 7 (Tailwind) - Backend: Node.js + Express 5 - ORM: Prisma (with a", "context": "maker", "use_case": "translating rules to JSON Logic", "switched_from": "", "story": "StackForge \u2013 Visual full-stack generator using JSON Logic"}, {"id": 742962721, "item_id": 46435170, "url": "https://news.ycombinator.com/item?id=46435170", "by": "ImPrajyoth", "date": "2025-12-30", "stance": "neutral", "quote": "With Groq speed (Llama 3 @ 800t/s)", "excerpt": "OP here. this is a cursed project lol, but i wanted to see: What happens if you replace the OS scheduler with an LLM? With Groq speed (Llama 3 @ 800t/s), inference is finally fast enough to be in the system loop. i built this TUI to monitor my process tree. instead of just showing CPU %, it checks the context (parent process, disk I/O) to decide if a process is compiling code or bloatware. It roasts, throttles, or ki", "context": "", "use_case": "classifying processes", "switched_from": "", "story": "Show HN: Replacing my OS process scheduler with an LLM"}, {"id": 741096464, "item_id": 46318529, "url": "https://news.ycombinator.com/item?id=46318529", "by": "eibrahim", "date": "2025-12-18", "stance": "warn", "quote": "I tried llama a few months back and the output was very disappointing.", "excerpt": "I tried llama a few months back and the output was very disappointing. I don't think it will handle book writing well or produce good quality. what do you mean tts for interviews? You mean use AI to interview grandpa and get his story and then use that to create \"his story\"?", "context": "", "use_case": "book writing", "switched_from": "", "story": "Ask HN: How would you monetize an AI book-writing app?"}, {"id": 735736308, "item_id": 45983519, "url": "https://news.ycombinator.com/item?id=45983519", "by": "mesadb", "date": "2025-11-19", "stance": "warn", "quote": "It looked nice but the latency was not acceptable for everyday use.", "excerpt": "Some implementation details, since getting this to work well was not trivial. My goal was \u201cpress hotkey, start talking, see text within ~1\u20132 seconds\u201d on an M2 MacBook Pro, and support multiple languages. First attempts (cloud) \u2013 I tried Hugging Face real-time transcription. It worked but latency was all over the place and costs would not scale. \u2013 I tried OpenAI real-time transcription. Latency was better, but when th", "context": "built a macOS dictation app", "use_case": "formatting transcripts into email", "switched_from": "", "story": "Show HN: Slave \u2013 local dictation and TTS for macOS (3k words free)"}, {"id": 732554066, "item_id": 45784629, "url": "https://news.ycombinator.com/item?id=45784629", "by": "Weloveall", "date": "2025-11-01", "stance": "neutral", "quote": "I use ollama in openwebui and llama3 in model.", "excerpt": "I use ollama in openwebui and llama3 in model. But I also have an ssh flask where I use an claude api.", "context": "", "use_case": "local LLM", "switched_from": "", "story": "Ask HN: Who uses open LLMs and coding assistants locally? Share setup and laptop"}]}