--- title: What AI can my computer run? canonical: "https://consideredgear.com/tools/local-ai-fit/" description: "Find a sensible local AI starting point for your computer, get a first test to try, and see what to check before spending money on an upgrade." --- # What AI can my computer run? Find your first model and find out if you need to buy anything. [Open the guided tool](https://consideredgear.com/tools/local-ai-fit/). Choose chat/writing, document excerpts or coding help; your computer type and memory; and an optional USD spending preference. No account, no computer scanning and no model call. Choices stay in the page. ## Two starting plans you can use now These are example profiles, not a scan of your computer or a measured fit. Each starts with chat and writing using hardware you already own. [Install or update Ollama](https://ollama.com/download) and check its current system requirements first. ### 16 GB Apple silicon Mac First candidate: Qwen3.5 4B — exact tag qwen3.5:4b-q4_K_M. Apple silicon Mac · 16 GB system unified memory. 3.3 GB is the published download size; runtime memory is higher. After installing Ollama, open Terminal or PowerShell and run: ```text ollama run qwen3.5:4b-q4_K_M ``` Inside the chat, enter: ```text /set parameter num_ctx 4096 ``` Try: Rewrite this as a friendly two-sentence message. Keep the date and time unchanged: Our appointment is on Tuesday at 3 pm. Please bring your laptop. Check: It should keep Tuesday and 3 pm, mention the laptop, and use two sentences. Then try one of your own everyday writing tasks. Before buying: Use what you own. If the first try is slow or runs out of memory, reduce context or test a smaller model before spending. ### RTX 3060 12 GB PC with 16 GB RAM First candidate: Qwen3.5 9B — exact tag qwen3.5:9b-q4_K_M. PC with an NVIDIA RTX GPU · 16 GB system RAM · 12 GB dedicated graphics memory. 6.6 GB is the published download size; runtime memory is higher. After installing Ollama, open Terminal or PowerShell and run: ```text ollama run qwen3.5:9b-q4_K_M ``` Inside the chat, enter: ```text /set parameter num_ctx 4096 ``` Try: Rewrite this as a friendly two-sentence message. Keep the date and time unchanged: Our appointment is on Tuesday at 3 pm. Please bring your laptop. Check: It should keep Tuesday and 3 pm, mention the laptop, and use two sentences. Then try one of your own everyday writing tasks. Before buying: Use what you own. If the first try is slow or runs out of memory, reduce context or test a smaller model before spending. These are starting experiments, not guaranteed results. Free memory, drivers and other applications affect performance. See the selection method and sources below, checked 2026-10-18. ## Starting points, not fit guarantees Apple silicon: 2B at 8 GB, 4B at 16 GB, 9B at 24 GB or more unified memory. NVIDIA RTX PCs with at least 16 GB system RAM: 4B at 6 GB dedicated VRAM for chat (8 GB for document/code trials), 9B at 12 GB or more VRAM. Otherwise a 2B trial with at least 8 GB RAM; other/older computers get a CPU trial with unconfirmed GPU acceleration. Below 8 GB system RAM or with unknown required memory, get more information or try a different machine. System RAM and VRAM are not added together. These are conservative editorial capacity screens, not measured free-memory requirements or benchmarks. Download size is not runtime memory. Check current OS/driver support. Start at 4096 tokens for chat or 8192 tokens for document/code trials. Images, whole repositories and entire document libraries are outside the fit estimates. Long texts need extraction/retrieval and a suitable application. - Qwen3.5 2B: exact tag qwen3.5:2b-q8_0, published download about 2.7 GB. [Model details](https://ollama.com/library/qwen3.5:2b-q8_0). - Qwen3.5 4B: exact tag qwen3.5:4b-q4_K_M, published download about 3.3 GB. [Model details](https://ollama.com/library/qwen3.5:4b-q4_K_M). - Qwen3.5 9B: exact tag qwen3.5:9b-q4_K_M, published download about 6.6 GB. [Model details](https://ollama.com/library/qwen3.5:9b-q4_K_M). ## Try a useful task before spending ### Chat & writing Prompt: Rewrite this as a friendly two-sentence message. Keep the date and time unchanged: Our appointment is on Tuesday at 3 pm. Please bring your laptop. Check: It should keep Tuesday and 3 pm, mention the laptop, and use two sentences. Then try one of your own everyday writing tasks. ### Work with documents Prompt: Use only this note. Return the owner, due date and budget. If a detail is missing, say not stated. Note: Maya will deliver the draft by Friday. Budget: $240. Who approved it? Check: Expect Maya, Friday and $240. The approver is not stated. Start with pasted text; a folder of PDFs needs a separate document app, extraction and retrieval. ### Help with code Prompt: Explain this Python function, identify the empty-input problem, and suggest a fix: def average(values): return sum(values) / len(values) Check: It should identify division by zero for an empty list and explain its chosen behavior. Review and test generated code before using it. Start with a function, not a whole repository. Install/update Ollama from its official site, download the selected exact model, set the starting context and check an actual task. The terminal command is ollama run followed by the selected model tag. Inside the chat use /set parameter num_ctx with the indicated context. In another terminal, ollama ps shows CPU/GPU placement. Judge answers and response time before buying hardware; review and test generated code. ## Buying guidance The guide recommends trying existing hardware first. A spending preference never increases your computer's capability. If a trial shows a memory limit, compare compatible upgrades and the full delivered system cost. Confirm case space, power, connectors, cooling and whether RAM is upgradeable. Unknown prices stay unknown: no product is claimed to fit a budget without a current quote. [Model memory](https://consideredgear.com/guides/how-much-vram-local-llm/) · [Hardware buying guide](https://consideredgear.com/guides/best-way-to-run-ai-at-home-2026/) · [Check ownership cost vs a monthly AI bill](https://consideredgear.com/tools/local-ai-cost-calculator/). ## Sources and method Checked 2026-10-18. The model family is a reproducible starting point, not a universal quality ranking. Your hardware has not been benchmarked by this guide. - [Ollama hardware support](https://docs.ollama.com/gpu) - [Ollama installation and first run](https://docs.ollama.com/quickstart) - [Context length and memory](https://docs.ollama.com/context-length) - [Set context and inspect GPU use](https://docs.ollama.com/faq) - [Qwen3.5 2B exact download](https://ollama.com/library/qwen3.5:2b-q8_0) - [Qwen3.5 4B exact download](https://ollama.com/library/qwen3.5:4b-q4_K_M) - [Qwen3.5 9B exact download](https://ollama.com/library/qwen3.5:9b-q4_K_M)