On this page
The recommendationWhy this model is a sensible first trySet a small first contextWhat community tests can and cannot tell youVerify it on your own computerSourcesStart with Qwen3 14B Q4_K_M on a 24 GB CPU system. The 9.3 GB file leaves meaningful room for the OS and context; test larger models only after measuring free memory and speed.
Here is what matters, why it matters and how to check it on your own setup.
The recommendation
Start with Qwen3 14B in the exact Ollama Q4_K_M artifact shown below. Its published download is about 9.3 GB; this is the file size, not the full running-memory requirement. The command uses a current named Ollama tag so you can verify the model name and size before you download it.
No model is guaranteed to run on every computer with the same advertised memory. Operating system use, integrated graphics, context length, other apps, runtime version and hardware support all change the available headroom. This is a conservative first candidate, not a claim that we personally tested your machine.
Run this exact Ollama model
ollama run qwen3:14b-q4_K_MCheck that the exact tag is available in the Ollama library for your operating system. Stop if your free memory is below the model file size plus the system and context headroom described here.
Why this model is a sensible first try
The Ollama library publishes the model family, quantized artifact and download size. Qwen's model documentation describes the family and supported tasks. These sources confirm that the named artifact exists; they do not prove a particular speed on your computer.
A 4-bit Q4_K_M artifact is smaller than the full-precision weights, but the running process also needs memory for context, temporary buffers and the operating system. Keep the first test short and avoid opening many large apps at once.
Set a small first context
Begin with a 4,096-token context (often shown as 4K) in your app. Context is the text the model can consider in one conversation. Longer context can use substantially more memory, and multiple simultaneous requests increase memory use.
Ollama's context settings depend on the version and available accelerator memory. Check the current official instructions rather than assuming the model card's maximum context is a sensible laptop setting.
What community tests can and cannot tell you
We found official model files and hardware-specific public tests, but no independent run that matches every computer in this memory tier. We therefore publish no invented tokens-per-second figure and do not call this a guaranteed fit. Use the repeatable check below on your device.
Verify it on your own computer
- Check total RAM or VRAM and current free memory before the first launch.
- Start the model with other heavy applications closed and a short 4K context.
- Ask it to write a short email, summarize a small passage and solve one real task you care about.
- Watch memory use and note load failures, disk swapping, freezing, first-token delay and answer speed.
- Repeat with the apps you normally keep open. If it slows or fails, step down one model size or context level.
- Do not use its answer for consequential work without checking it against trusted sources.
Words you may see in this guide
- RAM
- The working memory used by your computer and apps.
- VRAM
- The memory on a dedicated graphics card. Shared graphics may use system RAM.
- Context
- The text and other input the model can use for its current answer.
- Quantization
- A way to make model weights smaller. It can change memory use and answer quality.
- Engine
- The software that loads the model and runs it.
- RAG
- A workflow that finds useful document passages before asking a model to answer.
- Benchmark
- A repeatable test. A result is useful only when you know the setup and task.
Official facts and real user evidence
Official documentation supports product and model facts. Community discussions show real setups, failures, and questions. A community result is supporting evidence, not a promise that another computer will perform the same way.
Watch a real test. Check what fits your computer.
See practical local AI tests from DeviceTerra, then use MamiLens to build a hardware-aware shortlist for your own setup.
This guide is educational. Model software, licenses, and hardware support can change. Check official sources before an important deployment.
