{"slug": "qwen2-5", "name": "Qwen2.5", "category": "ai", "category_name": "AI models and assistants", "description": "", "recommenders": 2, "commenters": 6, "warners": 1, "score": 1.519, "rising": -0.65, "first": "2026-01-14", "latest": "2026-09-26", "why": "I chose the 7B model because I wanted something that could run affordably at scale while still being capable enough for nuanced generation.", "why_is_quote": true, "has_verdict": false, "sentiment": {"recommend": 2, "neutral": 4, "compare": 0, "warn": 1}, "spark": [0, 0, 0, 2, 0, 0, 0, 0, 0, 0, 0, 0], "listed": false, "loved": [], "caveat": "", "short_quote": "I chose the 7B model because I wanted something that could run affordably at scale while still being capable enough for nuanced generation.", "icon": false, "domain": null, "github_repo": null, "official_url": "", "pricing": "", "aliases": ["Qwen2.5", "Qwen2.5-7B-Instruct", "Qwen 2.5"], "verdict": null, "praise": [{"label": "", "summary": "", "quotes": [{"id": 751305377, "item_id": 46956586, "url": "https://news.ycombinator.com/item?id=46956586", "by": "0xbadcafebee", "date": "2026-02-09", "stance": "recommend", "quote": "I use smaller models (qwen2.5:1.5b) with keyboard shortcuts and speech to text to ask for man page entries", "excerpt": "Local models are decent now. Qwen3 coder is pretty good and decent speed. I use smaller models (qwen2.5:1.5b) with keyboard shortcuts and speech to text to ask for man page entries, and get 'em back faster than my internet connection and a \"robust\" frontier model does. And web search/RAG hides a multitude of sins. \"Using anything other than the frontier models is actively harmful\" - so how come I'm getting solid resu", "context": "", "use_case": "man page lookups", "switched_from": "", "story": "Eight More Months of Agents"}, {"id": 750936450, "item_id": 46933528, "url": "https://news.ycombinator.com/item?id=46933528", "by": "graphpilled", "date": "2026-02-08", "stance": "recommend", "quote": "I chose the 7B model because I wanted something that could run affordably at scale while still being capable enough for nuanced generation.", "excerpt": "Hi HN, I'm a computer systems engineering student in Mexico who switched from film school. I built CineGraphs because my filmmaker friends and I kept hitting the same wall\u2014we'd have a vague idea for a film but no structured way to explore where it could go. Every AI writing tool we tried output generic, formulaic slop. I didn't want to build another ChatGPT wrapper, so I went a different route. The idea is simple: yo", "context": "", "use_case": "fine-tuned screenplay generation", "switched_from": "", "story": "Show HN: Fine-tuned Qwen2.5-7B on 100 films for probabilistic story graphs"}]}], "caveats": [{"label": "", "summary": "", "quotes": [{"id": 797370978, "item_id": 49835686, "url": "https://news.ycombinator.com/item?id=49835686", "by": "jpusateri", "date": "2026-09-24", "stance": "warn", "quote": "Adding a 7B generative LLM reranker (Qwen2.5-7B-Instruct) downstream degraded top-1 accuracy", "excerpt": "Author here. In building compliance automation at Middle Coast Software, one of the biggest friction points was mapping technical vulnerability scans (CVEs) to formal security control catalogs (NIST SP 800-53 Rev 5 and CMMC). The problem is a severe vocabulary gap: vulnerability reports describe low-level exploit mechanics (buffer overflows, memory corruption, protocol stacks), while compliance controls describe admi", "context": "", "use_case": "reranking", "switched_from": "", "story": "Show HN: Fine-tuned 110M encoder beat 7B LLMs and hybrid search for NIST mapping"}]}], "who": [{"by": "dsjakupov", "context": "built a privacy-focused app", "url": "https://news.ycombinator.com/item?id=49859219", "date": "2026-09-26"}], "tags": [], "use_cases": ["on-device private diary app", "reranking", "local model on raspberry pi", "man page lookups", "fine-tuned screenplay generation", "local task routing"], "compared": [{"slug": "qwen3-vl", "name": "Qwen3-VL", "co": 2, "switched_from": 0, "switched_to": 0}], "makers": [], "threads": [{"url": "https://news.ycombinator.com/item?id=49807413", "title": "Ask HN: When is fine-tuning a small LLM worth it?", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49835684", "title": "Show HN: Fine-tuned 110M encoder beat 7B LLMs and hybrid search for NIST mapping", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=47365061", "title": "Show HN: OpenLight \u2013 Lightweight Telegram AI Agent for Raspberry Pi", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=46933223", "title": "Eight More Months of Agents", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=46933515", "title": "Show HN: Fine-tuned Qwen2.5-7B on 100 films for probabilistic story graphs", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=46743025", "title": "Show HN: RWS \u2013 Local-first task orchestrator using Qwen 2.5 on consumer hardware", "mentions": 1}], "sentiment_by_month": {"2025-11": {}, "2025-12": {}, "2026-01": {"neutral": 2}, "2026-02": {"recommend": 2}, "2026-03": {"neutral": 1}, "2026-04": {}, "2026-05": {}, "2026-06": {}, "2026-07": {}, "2026-08": {}, "2026-09": {"warn": 1, "neutral": 1}, "2026-10": {}}, "quotes": [{"id": 797747504, "item_id": 49859219, "url": "https://news.ycombinator.com/item?id=49859219", "by": "dsjakupov", "date": "2026-09-26", "stance": "neutral", "quote": "I used qwen2.5 for the mobile use", "excerpt": "Yeah, I did it for my app that specializes in privacy. Basically I used qwen2.5 for the mobile use and the case I was solving is user's most private thoughts - smth like a private diary But I am thinking about how can I fine-tune the model on the users' data and keep them private? Maybe like improve the model directly on the phone ahhaha", "context": "built a privacy-focused app", "use_case": "on-device private diary app", "switched_from": "", "story": "Ask HN: When is fine-tuning a small LLM worth it?"}, {"id": 797370978, "item_id": 49835686, "url": "https://news.ycombinator.com/item?id=49835686", "by": "jpusateri", "date": "2026-09-24", "stance": "warn", "quote": "Adding a 7B generative LLM reranker (Qwen2.5-7B-Instruct) downstream degraded top-1 accuracy", "excerpt": "Author here. In building compliance automation at Middle Coast Software, one of the biggest friction points was mapping technical vulnerability scans (CVEs) to formal security control catalogs (NIST SP 800-53 Rev 5 and CMMC). The problem is a severe vocabulary gap: vulnerability reports describe low-level exploit mechanics (buffer overflows, memory corruption, protocol stacks), while compliance controls describe admi", "context": "", "use_case": "reranking", "switched_from": "", "story": "Show HN: Fine-tuned 110M encoder beat 7B LLMs and hybrid search for NIST mapping"}, {"id": 757841058, "item_id": 47365066, "url": "https://news.ycombinator.com/item?id=47365066", "by": "evgenii_isupov", "date": "2026-03-13", "stance": "neutral", "quote": "currently using qwen2.5:0.5b", "excerpt": "Built this mainly as a lightweight alternative to heavier AI agent frameworks. Runs on Raspberry Pi 5 with Ollama (currently using qwen2.5:0.5b). Telegram acts as the control interface for system monitoring, services and notes.", "context": "", "use_case": "local model on Raspberry Pi", "switched_from": "", "story": "Show HN: OpenLight \u2013 Lightweight Telegram AI Agent for Raspberry Pi"}, {"id": 751305377, "item_id": 46956586, "url": "https://news.ycombinator.com/item?id=46956586", "by": "0xbadcafebee", "date": "2026-02-09", "stance": "recommend", "quote": "I use smaller models (qwen2.5:1.5b) with keyboard shortcuts and speech to text to ask for man page entries", "excerpt": "Local models are decent now. Qwen3 coder is pretty good and decent speed. I use smaller models (qwen2.5:1.5b) with keyboard shortcuts and speech to text to ask for man page entries, and get 'em back faster than my internet connection and a \"robust\" frontier model does. And web search/RAG hides a multitude of sins. \"Using anything other than the frontier models is actively harmful\" - so how come I'm getting solid resu", "context": "", "use_case": "man page lookups", "switched_from": "", "story": "Eight More Months of Agents"}, {"id": 750936450, "item_id": 46933528, "url": "https://news.ycombinator.com/item?id=46933528", "by": "graphpilled", "date": "2026-02-08", "stance": "recommend", "quote": "I chose the 7B model because I wanted something that could run affordably at scale while still being capable enough for nuanced generation.", "excerpt": "Hi HN, I'm a computer systems engineering student in Mexico who switched from film school. I built CineGraphs because my filmmaker friends and I kept hitting the same wall\u2014we'd have a vague idea for a film but no structured way to explore where it could go. Every AI writing tool we tried output generic, formulaic slop. I didn't want to build another ChatGPT wrapper, so I went a different route. The idea is simple: yo", "context": "", "use_case": "fine-tuned screenplay generation", "switched_from": "", "story": "Show HN: Fine-tuned Qwen2.5-7B on 100 films for probabilistic story graphs"}, {"id": 747888625, "item_id": 46743039, "url": "https://news.ycombinator.com/item?id=46743039", "by": "resworksentinel", "date": "2026-01-24", "stance": "neutral", "quote": "It uses Qwen 2.5 and handles urgency detection and routing without any API calls.", "excerpt": "Hi HN, I built this because I wanted a privacy-focused alternative to LangChain/Zapier that runs entirely on my RTX 3080. It uses Qwen 2.5 and handles urgency detection and routing without any API calls. Here is a short demo video of it handling 500+ tasks: https://youtu.be/tky3eURLzWo The code is pure Python (no heavy frameworks). Feedback on the routing logic is welcome.", "context": "", "use_case": "local task routing", "switched_from": "", "story": "Show HN: RWS \u2013 Local-first task orchestrator using Qwen 2.5 on consumer hardware"}, {"id": 746059521, "item_id": 46628720, "url": "https://news.ycombinator.com/item?id=46628720", "by": "graphpilled", "date": "2026-01-14", "stance": "neutral", "quote": "fine-tuned Qwen2.5-7B-Instruct with a LoRA optimized for probabilistic story branching", "excerpt": "Hi HN, I'm a computer systems engineering student in Mexico who switched from film school. I built CineGraphs because my filmmaker friends and I kept hitting the same wall\u2014we'd have a vague idea for a film but no structured way to explore where it could go. Every AI writing tool we tried output generic, formulaic slop. I didn't want to build another ChatGPT wrapper, so I went a different route. The idea is simple: yo", "context": "", "use_case": "fine-tuned story model", "switched_from": "", "story": "Show HN: Fine-tuned Qwen2.5-7B on 100 films for probabilistic story graphs"}]}