34 articles · newest first
8GB gets E4B, 12GB gets 12B, 16GB gets 12B at Q8 (or 26B-A4B with Google's own 4-bit file), and 24GB gets 26B-A4B or 31B. The VRAM for each size and quant, the exact Ollama tag (the bare one pulls a small model), the 4K-context default that catches Ollama users on cards under 24GB, and when Qwen3.8 27B is the better pick.
Aliteq Team · · 13 min
The GitHub request to keep MoE experts in RAM is still open after 15 months. But since Ollama 0.30, llama.cpp's own fit step can park them there for you, as long as you leave num_gpu alone. Plus what Ollama's Qwen3.8 Flash Next tags need, and why Reddit can tell when you're running Ollama.
Aliteq Team · · 8 min
OpenCode only works when the model can call tools and holds 64K of context. Seven local models pass the first test. Here is what each one needs in VRAM, worked out from their configs.
Aliteq Team · · 11 min
One person on one machine, or a team hitting one server? That decides it. Then licence, hardware, ports, security defaults, and where Strata fits. Read Nov 1, 2026.
Aliteq Team · · 10 min
Qwen2.5-Coder-32B still runs a serious private coding assistant on a single 24GB card, with a 32K native context (128K with YaRN). The VRAM math, the tight-but-real fit, and how it stands against the newer Qwen3-Coder.
Aliteq Team · · 7 min
Microsoft's Phi-4 is a compact 14B that runs on a mid-range card — but at 16K context and a late-2024 release, it shows its age next to newer 14B-class models. The VRAM math and an honest verdict.
Aliteq Team · · 7 min
Qwen3.6's dense 27B is the no-fuss 24GB-card option — every parameter fires, nothing to offload, a real 256K context. The VRAM math, the GPU, and how it differs from the 35B-A3B MoE.
Aliteq Team · · 7 min
Qwen3-Next 80B-A3B activates only 3B parameters per token — so it's fast. But 80B of weights still have to live somewhere. The honest VRAM math, the offload trick, and what it really takes.
Aliteq Team · · 7 min
The fast 35B MoE that thinks like a 3B is now last generation. It still makes sense for speed on a 16 to 24 GB card and for its coder fine-tunes. For the best answers, Reddit owners are moving to Qwen3.8 27B or Flash Next.
Aliteq Team · · 8 min
Type ollama run qwen3.5 and this is the model you get — the 9B. What it is, how little VRAM it needs, and the cheapest card that runs it well.
Aliteq Team · · 7 min
The same kind of model can run on a company's servers or on your own computer. What flips for privacy, cost, speed and model choice — and why your GPU's memory is the ceiling.
Aliteq Team · · 4 min
For most small teams it's Open WebUI: one container, roles, SSO and free to use as long as its branding stays on screen. LibreChat if you must rebrand it, AnythingLLM if the team mostly chats with documents or wants it hosted. What each one costs, needs and allows, from their own licences and docs read Nov 3, 2026.
Aliteq Team · · 9 min
LM Studio is a desktop app that finds, downloads and runs models for you. Open WebUI is a web platform you host for yourself, your family or a team. Here's when each one wins, what changed this week, and the five-minute setup that uses both.
Aliteq Team · · 7 min
Yes, as long as the Open WebUI name and logo stay exactly as shipped. The 50-user number in its license only limits rebranding. Here's what Open WebUI's own license and docs say about logos, white-labeling, client hosting and what does need a paid license, read Nov 3, 2026.
Aliteq Team · · 7 min
Ollama's default gemma4 is the small E4B and qwen3.5 is the 9B, not what most people assume. What each tag really downloads, what the bigger models need at the 64K context agents use, and what to pick for 12, 16 and 24GB cards.
Aliteq Team · · 7 min
The two biggest open-source personal agents, compared on what their docs actually say: licence, chat apps, skills, safety defaults, and the one-command Ollama setup for each.
Aliteq Team · · 7 min
An agent that reads your messages and runs your tools can be talked into things. That's the core risk, and it's what OpenClaw's own advisories show. The plain-English version, and the safe setup step by step.
Aliteq Team · · 8 min
All three coding agents can run on your own GPU through Ollama. The catch is 64k of context, which Ollama doesn't give you by default. Which agent to pick, and what your card can actually hold.
Aliteq Team · · 9 min
ollama launch opencode or one opencode.json block, plus the 64k context OpenCode requires, why tool calls fail without it, and what fits on your GPU at that length.
Aliteq Team · · 6 min
ollama launch codex, the --oss flag, or a permanent profile, plus the 64k context Codex needs and what gpt-oss 20B vs 120B costs in VRAM at that length.
Aliteq Team · · 6 min
One command, or three environment variables. Plus the 64k context setting Ollama's defaults don't give you, the VRAM that costs, and what Ollama's Anthropic API doesn't support.
Aliteq Team · · 6 min
Yes, Arc runs local LLMs well now — but skip the old IPEX-LLM guides (Intel archived it in January 2026). Here's the current, working path: Ollama or LM Studio on the Vulkan backend, step by step.
Aliteq Team · · 7 min
Quantization is the trick that lets a 16GB card run models it has no business running. Here's how GGUF quant levels work, which one I default to, the memory math, and the honest quality trade — so you get the most model your VRAM can hold.
Aliteq Team · · 9 min
AMD went from 'buy Nvidia unless you enjoy pain' to 'this just works' in 2026. The two setup paths — ROCm for speed, Vulkan for ease — which to pick, and how to get a model on your Radeon's GPU tonight.
Aliteq Team · · 8 min
No graphics card comes close. DeepSeek V4.1 Flash is a 763-billion-parameter download that needs a 4x H200 or 8x H100 server, a 512GB Mac Studio, or Unsloth's 1-bit build on 128GB+ of RAM. What each route costs, what Ollama offers, how it differs from V4 Flash, and when DeepSeek's API is cheaper, checked Nov 7, 2026.
Aliteq Team · · 18 min
Uncensored local models answer without the refusals and lectures of cloud chatbots. How the two kinds are made, how to run one in Ollama on an 8 GB card, and the rules that still apply after the filter is gone.
Aliteq Team · · 8 min
Mistral's models are famously efficient — they punch above their size, so you get a lot on modest hardware. Here's how to run Mistral locally in one command, by size.
Aliteq Team · · 10 min
Qwen3.8 27B is the default pick for a 16 to 24GB card, and Flash Next runs on smaller cards with lots of RAM. One command gets you started. Three default settings decide whether it is any good.
Aliteq Team · · 9 min
Only the distilled DeepSeek-R1 models fit a home PC: the 8B on an 8GB card, the 14B on 12 to 16GB, the 32B on 24GB, the 70B on 48GB. DeepSeek's newest V4 models are 284B to 1.6T parameters, and the DeepSeek models Ollama lists as cloud-only don't run on your machine. Install steps, sizes and the one setting that trips up 24GB cards, checked Nov 3, 2026.
Aliteq Team · · 8 min
Both are free and run the same open models. Pick LM Studio to browse and tune models in an app, Ollama for a model your other tools call. Checked Nov 1, 2026.
Aliteq Team · · 9 min
Running your own AI model at home sounds technical. It isn't — you can be chatting with a private, offline AI in about five minutes with one free app. Here's exactly how.
Aliteq Team · · 9 min
The Arc B580's 12GB holds every 8B model and 14B models at 4-bit, and Ollama now runs on it through Vulkan. The catch in November 2026 is the price: about $378 on eBay against a $249 list price, while a used RTX 3060 12GB with CUDA averaged $322.
Aliteq Team · · 8 min
Short answer: Qwen3.8 27B. On our VRAM numbers it fits a 24GB card at Q5_K_M with 32K of context, and Ollama's default download is 18 GB. Gemma 4 and Qwen3.6 MoE models are faster, gpt-oss-20b leaves the most room, and NVIDIA's 24GB Nemotron comes with a runtime catch.
Aliteq Team · · 9 min
Your VRAM picks the model before you do. What fits on 8, 12, 16, 24 and 32GB cards, worked out on our VRAM engine and checked against each model card and Ollama's download sizes on Nov 6, 2026, plus why NVIDIA's "24GB" Nemotron wants a 32GB card in llama.cpp.
Aliteq Team · · 8 min