gpt-oss

7 articles · newest first

Flat illustration of a person at a desk with a small plain mini desktop computer and an AI brain shape split into two pieces above it

gpt-oss-20b on a Mac mini (16GB): yes, but get 24GB

llama.cpp's own guide runs it on a 16GB Mac with one flag. But on a Mac the CPU and GPU share one pool of memory, so the whole model still has to fit, and 16GB leaves almost nothing for macOS. Here's the maths.

Aliteq Team · Updated Oct 4, 2026 · 6 min

Flat illustration of a person splitting an AI brain puzzle piece between a small graphics card and a row of memory sticks

gpt-oss-20b on an 8GB or 12GB GPU: yes, with --n-cpu-moe

Everyone says 16GB minimum. That's only true if the whole model has to live on the GPU. llama.cpp's own guide runs it on an 8GB RTX 2060. Here are the commands, and the starting values for your card.

Aliteq Team · Sep 25, 2026 · 7 min

Flat illustration of two people on a sofa looking at a laptop together, with a small glowing box on the side table

Software

How to use Ollama with Codex CLI (local gpt-oss, 2026)

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 · Updated Sep 26, 2026 · 6 min

A server GPU representing large model inference

gpt-oss-120b hardware requirements: about 64 GB of memory, and why a smaller quant won't save you

OpenAI's 117B open model ships its experts in 4-bit MXFP4, so the files are about 63 GB whichever quant you pick. Here is how much VRAM and RAM it needs at 8K to 128K context, which GPUs and machines hold it, and when a 24 GB card plus RAM is enough.

Aliteq Team · Updated Oct 29, 2026 · 9 min

Night line-art illustration next to the headline: 16GB just got bigger From gpt-oss-20b to Qwen3.8 27B. A card reads gpt-oss-20b on 16GB: 32K context (our VRAM engine).

16GB GPU? Which local AI models fit now, from gpt-oss-20b to Qwen3.8 27B (Nov 2026)

16GB used to stop at 14B models. Now gpt-oss-20b fits with 32K of context, and Qwen3.8 27B fits at a 3-bit quant. Here is what runs on an RTX 5060 Ti, 4060 Ti or 5070 Ti, on our numbers, and when system RAM takes you further.

Aliteq Team · Updated Nov 1, 2026 · 8 min

Best model for 24GB: 3090 or 4090? Start with Qwen3.8 27B. A screenshot of Ollama's qwen3.8 tag list with the 27b q4_K_M row outlined, and a card reading qwen3.8:27b Ollama download 18 GB, read Nov 6, 2026.

What's the best model to run on a 24GB GPU like a 3090 or 4090? Qwen3.8 27B, and when to pick another

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 · Updated Nov 6, 2026 · 9 min

Line drawing of a graphics card next to the headline Which model fits your VRAM? 8, 12, 16, 24 and 32GB cards, model by model, with a card reading 24GB card, Qwen3.8 27B at Q5_K_M with 32K context: 21.2 GB on our VRAM engine, Nov 6, 2026

Which local AI model fits my GPU's VRAM? The 8GB to 32GB chart, model by model

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 · Updated Nov 6, 2026 · 8 min