Qwen3 14B
The point where a local model starts to feel smart. Noticeably better than 8B at reasoning and code, and it still fits a 12 GB card.
ollama run qwen3:14bNeeds about 11 GB of graphics memory, or unified memory on a Mac. First time? Install Ollama, then paste that into a terminal.
Download9.3 GB
Memory to runAbout 11 GB
Read per token9.3 GB
LicenseApache 2.0
Context32K tokens, 128K with YaRN
ReleasedApril 2025
Good for
- Coding help and code review
- Reasoning through problems with thinking mode
- Tool calling and simple agents
- Best pick for 12 GB cards
Not the best pick for
- Cards with 8 GB or less: it won't fit with room for context
110 GPUs and Macs
What runs Qwen3 14B
Estimated speed on each card at the smallest memory size that fits it, from memory bandwidth. As a dense model it reads all its weights for every token, so memory bandwidth sets the speed.
Runs great 62 cards
- H200141 GB · ~280 tok/s
- H10080 GB · ~200 tok/s
- RTX 509032 GB · ~110 tok/s
- RTX Pro 6000 Blackwell96 GB · ~110 tok/s
- A10040 GB · ~92 tok/s
- RTX Pro 5000 Blackwell48 GB · ~79 tok/s
- Instinct MI10032 GB · ~73 tok/s
- Instinct MI5016 GB · ~61 tok/s
- RTX 3090 Ti24 GB · ~60 tok/s
- RTX 409024 GB · ~60 tok/s
- RTX 508016 GB · ~57 tok/s
- RTX 6000 Ada48 GB · ~57 tok/s
- RX 7900 XTX24 GB · ~57 tok/s
- RTX 309024 GB · ~55 tok/s
- A3024 GB · ~55 tok/s
- RTX 3080 Ti12 GB · ~54 tok/s
- Tesla V10016 GB · ~53 tok/s
- RTX 5070 Ti16 GB · ~53 tok/s
- RTX 5080 Laptop16 GB · ~53 tok/s
- RTX 5090 Laptop24 GB · ~53 tok/s
- RTX Pro 4500 Blackwell32 GB · ~53 tok/s
- L40S48 GB · ~51 tok/s
- Radeon Pro W790048 GB · ~51 tok/s
- Apple M3 Ultra96 GB · ~48 tok/s
- RX 7900 XT20 GB · ~47 tok/s
- Apple M1 Ultra64 GB · ~47 tok/s
- Apple M2 Ultra64 GB · ~47 tok/s
- RTX A500024 GB · ~45 tok/s
- RTX A550024 GB · ~45 tok/s
- RTX A600048 GB · ~45 tok/s
- RTX 308012 GB · ~45 tok/s
- RTX 4080 Super16 GB · ~44 tok/s
- Tesla P10012 GB · ~43 tok/s
- RTX 408016 GB · ~42 tok/s
- A4048 GB · ~41 tok/s
- RTX 4070 Ti Super16 GB · ~40 tok/s
- RTX 507012 GB · ~40 tok/s
- RTX 5070 Ti Laptop12 GB · ~40 tok/s
- Titan RTX24 GB · ~40 tok/s
- Quadro RTX 600024 GB · ~40 tok/s
- Quadro RTX 800048 GB · ~40 tok/s
- RTX Pro 4000 Blackwell24 GB · ~40 tok/s
- Titan V12 GB · ~39 tok/s
- RTX A450020 GB · ~38 tok/s
- RX 907016 GB · ~38 tok/s
- RX 9070 XT16 GB · ~38 tok/s
- Radeon AI Pro R970032 GB · ~38 tok/s
- RX 7800 XT16 GB · ~37 tok/s
- RTX 2080 Ti11 GB · ~36 tok/s
- A1024 GB · ~35 tok/s
- RTX 4090 Laptop16 GB · ~34 tok/s
- RTX 5000 Ada32 GB · ~34 tok/s
- RX 6950 XT16 GB · ~34 tok/s
- RX 7900 GRE16 GB · ~34 tok/s
- Radeon Pro W780032 GB · ~34 tok/s
- Arc A77016 GB · ~33 tok/s
- Apple M4 Max36 GB · ~32 tok/s
- RTX 3080 Ti Laptop16 GB · ~30 tok/s
- RX 680016 GB · ~30 tok/s
- RX 6800 XT16 GB · ~30 tok/s
- RX 6900 XT16 GB · ~30 tok/s
- Radeon Pro W680032 GB · ~30 tok/s
Usable 38 cards
- RTX 407012 GB · ~30 tok/s
- RTX 4070 Super12 GB · ~30 tok/s
- RTX 4070 Ti12 GB · ~30 tok/s
- GTX 1080 Ti11 GB · ~29 tok/s
- Apple M5 Max36 GB · ~27 tok/s
- Arc B58012 GB · ~27 tok/s
- Arc Pro B6024 GB · ~27 tok/s
- RTX 5060 Ti16 GB · ~26 tok/s
- RTX 3080 Laptop16 GB · ~26 tok/s
- Quadro RTX 500016 GB · ~26 tok/s
- RTX A400016 GB · ~26 tok/s
- RTX 4080 Laptop12 GB · ~26 tok/s
- RTX 4500 Ada24 GB · ~26 tok/s
- RX 7700 XT12 GB · ~26 tok/s
- Apple M1 Max32 GB · ~24 tok/s
- Apple M2 Max32 GB · ~24 tok/s
- Apple M3 Max36 GB · ~24 tok/s
- RX 6700 XT12 GB · ~23 tok/s
- RTX 306012 GB · ~21 tok/s
- RTX 4000 Ada20 GB · ~21 tok/s
- Tesla P4024 GB · ~20 tok/s
- RTX 206012 GB · ~20 tok/s
- Tesla T416 GB · ~19 tok/s
- RX 9060 XT16 GB · ~19 tok/s
- Apple M5 Pro24 GB · ~18 tok/s
- L424 GB · ~18 tok/s
- RTX 4060 Ti16 GB · ~17 tok/s
- RTX A200012 GB · ~17 tok/s
- Tesla M4012 GB · ~17 tok/s
- RX 7600 XT16 GB · ~17 tok/s
- Apple M4 Pro24 GB · ~16 tok/s
- NVIDIA DGX Spark128 GB · ~16 tok/s
- Ryzen AI Max+ 395 (Radeon 8060S)64 GB · ~15 tok/s
- Ryzen AI Max 390 (Radeon 8050S)32 GB · ~15 tok/s
- RTX 2000 Ada16 GB · ~13 tok/s
- Jetson AGX Orin32 GB · ~12 tok/s
- Apple M1 Pro16 GB · ~12 tok/s
- Apple M2 Pro16 GB · ~12 tok/s
Slow 10 cards
Cheapest build from our list that runs it well: Used desktop + RTX 3060 12 GB, from $600 used, at about 21 tok/s. See the builds.
Learn
Read more about Qwen3 14B
- A local coding assistant in VS Code: Continue, Cline and OllamaSet up a private AI coding assistant in VS Code with Ollama and Continue or Cline. Which model to pick for your card, and what to expect from it.
- AMD GPUs for local AI: ROCm, Vulkan and what works in practiceCan you run local LLMs on an AMD Radeon card? Which tools support it, ROCm vs Vulkan, the best AMD cards for AI, and the catches compared to NVIDIA.
- Building a quiet, always-on home AI serverHow to build a home AI server that's quiet enough to live with and cheap to leave on. Hardware picks, noise and power tips, and sharing it on your network.
- Chat with your documents locally, without uploading them anywhereAsk questions about your PDFs, notes and contracts with a model that runs on your own machine. How it works, how to set it up, and how to make the answers good.
- Does Ollama send your data anywhere? Local AI privacy, checkedWhat stays on your computer when you run a local model, what does go over the network, and how to check for yourself instead of taking anyone's word.
- Fix CUDA out of memory errors in Ollama and llama.cppWhy local models run out of GPU memory, and the fixes in order - smaller quant, shorter context, a quantized KV cache, fewer parallel slots, partial offload.
- How much VRAM do you need to run an LLM locally?A simple rule for how much graphics memory a model needs, what fits in 8, 12, 16, 24 and 48 GB, and why context length adds more.
- How to increase Ollama's context length (num_ctx)Ollama's default context is short and it cuts off long prompts quietly. Three ways to raise num_ctx, and how much memory a longer context costs.
- Intel Arc for local AI: can the B580 and A770 run LLMs?How Intel Arc cards run local LLMs, which tools support them, estimated speeds for the B580 and A770 16 GB, and the software catches to know first.
- LM Studio guide: download, run and serve local modelsHow to use LM Studio to find a model that fits your GPU or Mac, set GPU offload and context, and run its local OpenAI-compatible server.
- Local AI agents: tool calling, MCP and what works on small modelsHow agents and tool calling work with local models in Ollama, what MCP adds, which models handle tools reliably, and how to keep small models on track.
- Local AI for email: draft, summarize and sort mail privatelyUse a model on your own computer to draft replies, summarize threads and sort your inbox, without giving a cloud service your email. What works and what to avoid.
- Local AI for Home Assistant: a private voice assistant with OllamaConnect Home Assistant to a model running on your own hardware with the Ollama integration. What hardware you need, which model to pick, and how to set it up.
- Local AI for small business: a private assistant for your teamWhat a small business can do with AI running on its own hardware, what it costs, which build to buy, and how to share one machine with the whole team.
- Local embedding models: nomic-embed-text and friendsWhat embedding models do, why they power search over your documents, how to run nomic-embed-text with Ollama, and how to choose one.
- Local RAG that actually works: chunking, embeddings and rerankingWhy local document search gives bad answers and how to fix it. Practical settings for chunking, embedding models, hybrid search and reranking on your own machine.
- Local speech to text: Whisper model sizes and how to chooseRun Whisper on your own computer to transcribe audio privately. The model sizes, the speed and accuracy trade-off, and which tools to use.
- Ollama commands cheat sheet: every command you'll actually useThe Ollama commands for running, downloading, listing, stopping and deleting models, plus the in-chat slash commands and the settings worth knowing.
- Ollama using CPU instead of GPU? How to check and fix itOllama slow and your GPU idle? Check with ollama ps, then fix the usual causes - model too big, drivers, Docker, laptops, AMD and sleep - step by step.
- Qwen3 vs Gemma 3: which local model should you run?Qwen3 and Gemma 3 compared for local use. Reasoning, images, writing, licenses, memory and estimated speeds on 12 and 24 GB cards.
- Run a local LLM on Windows: setup with Ollama and an NVIDIA or AMD GPUInstall Ollama on Windows 10 or 11, get it using your graphics card, change where models are stored, and fix the common problems.
- Run Ollama in Docker, with NVIDIA GPU supportRun the official ollama/ollama Docker image on CPU or NVIDIA GPU. The NVIDIA Container Toolkit setup, Docker Compose, and checking the GPU is used.
- Serve a team on used GPUs: what our 90-card rack taught usHow to serve a local model to a team on cheap used GPUs. Our rack ran a billion tokens in 2.9 days on P100s, A4500s and mining cards. Builds and setup.
- Structured output with Ollama: get reliable JSON from a local modelUse Ollama's format parameter to make a local model return valid JSON, or JSON that matches your own schema. Working curl and Python examples, plus tips.
- Summarize meetings locally: transcribe and summarize without the cloudTurn a meeting recording into a transcript, summary and action items on your own computer with Whisper and a local model. Steps, models and the context-length catch.
- Temperature, top-p and min-p: the sampling settings that matterWhat temperature and the other sampling settings do, which ones are worth changing for local models, and how to set them in Ollama.
- The best GPU for local AI under $300The best graphics cards for running local LLMs on a $300 budget, what each one runs, estimated speeds, and which to pick for easy setup or raw value.
- The best laptop for local AI: what to look for and what to avoidHow to pick a laptop for running local LLMs. Why laptop GPU names mislead, how much VRAM you need, estimated speeds, and when a MacBook is the better choice.
- The best local LLM for 12 GB of VRAM (RTX 3060, 4070, B580)What to run on a 12 GB graphics card like the RTX 3060, 4070 or Arc B580. Picks for smarts, images and speed, with estimated speeds for each card.
- The best local LLM for 16 GB of VRAM (4060 Ti, 5060 Ti, 5080)What to run on a 16 GB card like the RTX 4060 Ti 16 GB, 5060 Ti, 5080, RX 9070 XT or a used Tesla P100, with estimated speeds and what doesn't fit.
- The best local LLM for a 16 GB Mac (MacBook Air, Mac mini)What runs well on a Mac with 16 GB of unified memory, how fast each M-series chip is, and how to keep the model from slowing your Mac down.
- The best local LLM for coding, by how much memory you haveWhich open models are worth using for code on 8, 12, 16, 24 GB and up, what they're good and bad at, and how to wire them into your editor.
- The best local LLM for reasoning and mathWhich open models reason best through math, logic and multi-step problems on 8 GB to 64 GB+, how thinking mode works, and why speed matters more than you'd expect.
- The best local LLM for writing, by how much memory you haveWhich open models write best for drafts, editing, emails and fiction on 8 GB to 48 GB+, plus the settings and prompts that make local writing sound less generic.
- The best local model for agents and tool callingWhich open models call tools reliably on 12 GB to 64 GB+, how tool calling works with Ollama, and what to expect from local agents compared with cloud models.
- The best local model for translationWhich open models translate best on your own hardware, by memory size, plus prompts that get natural results and where local translation still falls short.
- The best small local model for scripts, classification and JSONWhich small open model to use when a script calls it thousands of times to classify text, extract fields or return JSON, and how to get reliable structured output.
- The RTX 3060 12 GB for AI: the best cheap starter cardWhy the 12 GB RTX 3060 is the cheapest sensible GPU for local AI, what it runs and how fast, what to pay, and the 8 GB version to avoid.
- The Tesla P100 for local LLMs: 16 GB of HBM2 for under $100What we learned running a rack of 2016 data-center P100s for AI. Real speeds, what it takes to cool and power them, and the software catch.
- The Tesla P40 for local LLMs: 24 GB cheap, but is it fast enough?What the 24 GB Tesla P40 runs, how fast, how it compares to the P100, and the cooling, power and software catches before you buy one.
- Thinking models explained: reasoning mode, /think and effort levelsWhat reasoning mode does, when it's worth the wait, how to turn it on and off in Qwen3 and gpt-oss, and what it costs in speed and context.
- Tokens per second explained: what's a good speed for a local LLM?What tok/s means, how fast is fast enough for chat, coding and thinking models, and how to measure your own speed in Ollama.
- Uncensored local models: what abliterated means, and the tradeoffsWhy local models refuse some requests, what abliterated and uncensored fine-tunes change, what it costs in quality, and how to cut needless refusals without one.
- Use Ollama from other devices: OLLAMA_HOST, phones and other PCsMake Ollama reachable from your phone, laptop or another PC on your home network by setting OLLAMA_HOST on Mac, Windows and Linux, and keep it safe.
- Use your local AI from anywhere: Tailscale or Cloudflare TunnelReach Ollama or Open WebUI on your home GPU from your phone or laptop anywhere, without opening router ports. Tailscale vs Cloudflare Tunnel, set up.
- What to actually do with a local AI model: your first weekSeven practical things to try in your first week with a local model, from rewriting email to reading your own documents and scripting batch jobs.
Compare
Other models
- Qwen3 4BAbout 4 GB · Quick everyday chat and summaries
- Qwen3 8BAbout 7 GB · A solid all-rounder for chat and writing
- Gemma 3 12BAbout 10 GB · Reads images as well as text
- gpt-oss 20BAbout 16 GB · OpenAI's open model. Fast, strong reasoning
- Gemma 3 27BAbout 19 GB · Google's best open model for its size
- Qwen3 30B-A3BAbout 21 GB · 30B smarts at small-model speed
- Qwen3 32BAbout 22 GB · Strong coding and long answers
- Llama 3.3 70BAbout 45 GB · Close to the big cloud models
- gpt-oss 120BAbout 67 GB · OpenAI's largest open model
- Qwen3 235B-A22BAbout 144 GB · Alibaba's flagship open model, for multi-GPU rigs
- DeepSeek R1 671BAbout 406 GB · The full DeepSeek R1, for the biggest rigs