Reasoning20B official

gpt-oss

by OpenAI

gpt-oss: OpenAI's open-weight reasoning models. 20B runs in ~16 GB, 120B on one 80 GB card. Apache-2.0, adjustable reasoning effort.

#reasoning #tools #moe #agentic

arcflare pull ggml-org/gpt-oss-20b-GGUF

[ 01 ] // quickstart

┌─ arcflare — zsh // tty1

[ 02 ] // readme

gpt-oss

gpt-oss: OpenAI's open-weight reasoning models. 20B runs in ~16 GB, 120B on one 80 GB card. Apache-2.0, adjustable reasoning effort.

Get it

arcflare pull ggml-org/gpt-oss-20b-GGUF

arcflare run gpt-oss-20b

ArcFlare reads the GGUF header and loads it at the largest context your free VRAM can actually hold. Use arcflare fit to see that before downloading anything else, or arcflare use opencode to code with it.

Highlights

MoE with native MXFP4 weights: 21B total and 3.6B active (20b), 117B and 5.1B (120b)
Low, medium or high reasoning effort, set in the system prompt
Built for tool use and agents

Sizes

20b · 12GB
120b · 61GB

Other sizes live in sibling repos: swap the size in the repo name when pulling.

[ 03 ] // relatedmore →

glm4.5-air

Z.ai

Reasoning

GLM-4.5-Air: a 106B MoE with 12B active, hybrid reasoning and strong agent and coding scores, under MIT. Workstation-class.

size: 106B-A12B
vram: ~64 GB
license: MIT

2.7× a 24 GB card

#moe #agentic #reasoning

deepseek-r1-0528

DeepSeek

Reasoning

DeepSeek-R1-0528: the updated 671B reasoning model, plus an 8B distill on Qwen3 that runs on a gaming GPU. MIT.

size: 8B
vram: ~6 GB
license: MIT

25% of a 24 GB card

#reasoning #math #moe

mixtral

Mistral AI

Reasoning

Mixtral 8x7B — a sparse mixture-of-experts model offering strong quality at high inference efficiency.

size: 8x7B
vram: ~28 GB
license: Apache-2.0

1.2× a 24 GB card

#moe #reasoning #instruct