Hardware-aware · Open source · Aggiornato ogni settimana

AI locale: esegui modelli open source sul tuo computer.

L'hub per gli LLM locali: guide di installazione e un selettore hardware.

Puoi eseguire modelli AI come quelli dietro ChatGPT o Claude direttamente sul tuo portatile o desktop, senza inviare i tuoi dati da nessuna parte. RunLocal ti mostra quale modello open source scegliere per il tuo hardware, quale software gratuito installare (Ollama, LM Studio, llama.cpp), e come partire in circa dieci minuti. Nessuna conoscenza pregressa richiesta.

Trova un modello per il tuo hardware →Esegui il tuo primo modello Sfoglia la directory

Modelli che meritano il tuo disco

Una lista corta e con opinioni. Catalogo completo nella directory. Le schede sono in inglese.

Tutti i modelli →

Qwen 3.8

Alibaba (Qwen team) · China

2026

License: Apache 2.0 on the 27B; custom Qwen licence on the 2.4T MoE
Context: 262k tokens
Sizes: 27B dense · 2.4T-A95B MoE

CodingMultimodal workflowsMultilingualLocal workstations

August 2026, and the most-liked open-weight release of the summer. The 27B is dense, Apache 2.0, natively multimodal, and had GGUF builds from unsloth and bartowski within days — which makes it the current default for a 24 GB card. The 2.4T MoE sibling is frontier-scale and carries a different licence.

Official site →Run it locally

DeepSeek V4 Flash

DeepSeek AI · China

2026

License: MIT
Context: 1M tokens (65k native, extended 16x with YaRN)
Sizes: 304B-A13B MoE (284B before the DSpark decoder)

Agentic codingLong documentsHigh-memory workstations

The most-downloaded open-weight model on the Hub, and the largest entry here that a single machine can still load. 256 routed experts with 6 active per token means roughly 13B parameters fire on any given token, so it answers far faster than 304B suggests. The 0731 checkpoint supersedes the June preview and carries a DSpark speculative-decoding module, which is where the extra weights over the 284B base come from. Weights ship natively in FP8, so a Q8 build costs only about 7 GB more than Q4.

Official site →Run it locally

Qwen 3.6

Alibaba (Qwen team) · China

2026

License: Apache 2.0 on official open-weight checkpoints
Context: 262k native on 27B; long-context extensions available
Sizes: 27B dense · 35B-A3B MoE · additional large MoE variants

CodingMultimodal workflowsMultilingualLocal workstations

The previous open-weight Qwen generation, still widely deployed and the base for most community quantization work, including the 1-bit and ternary Bonsai builds. Qwen 3.8 supersedes it for new installs.

Official site →Run it locally

Qwen 3.5

Alibaba (Qwen team) · China

2026

License: Apache 2.0
Context: 262k native; selected variants support longer contexts
Sizes: 2B · 9B · 27B · 35B-A3B MoE · 122B-A10B MoE · 397B-A17B MoE

MultimodalMultilingualCodeCost-sensitive deployments

Released in 2026. The smaller checkpoints remain excellent local choices when Qwen 3.6 is too large.

Official site →Run it locally

Snapshot · Sep 14, 2026

In tendenza su Hugging Face

Browse HF directly →

Selezione automatica dall'Hugging Face Hub: un mix pesato di download, likes e freschezza. Si aggiorna ogni lunedì. Il badge di licenza è un'indicazione visiva, non un parere legale.

GLM-5.2

zai-org

Permissive

MIT

Downloads: 947.1k
Likes: 5.1k
Updated: 12 days ago

View on HF →score 0.87

DeepSeek-V4-Flash-0731

deepseek-ai

Permissive

MIT

Downloads: 4.5M
Likes: 4.0k
Updated: 1 months ago

View on HF →score 0.86

Qwen3.8-27B-Uncensored-GGUF

JonathanColetti · 27B

Permissive

Apache 2.0

Downloads: 2.9M
Likes: 1.1k
Updated: 16 days ago

View on HF →score 0.83

GLM-5.3

zai-org

Open weight

Custom license

Downloads: 712.6k
Likes: 1.8k
Updated: 10 days ago

View on HF →score 0.82

Qwen3.8-27B-OBLITERATED

OBLITERATUS · 27B

Permissive

Apache 2.0

Downloads: 1.2M
Likes: 1.2k
Updated: 20 days ago

View on HF →score 0.81

Ternary-Bonsai-27B-gguf

prism-ml · 27B

Permissive

Apache 2.0

Downloads: 644.3k
Likes: 1.3k
Updated: 13 days ago

View on HF →score 0.80

MiniCPM5-2B

openbmb · 2B

Permissive

Apache 2.0

Downloads: 206.8k
Likes: 1.4k
Updated: 2 days ago

View on HF →score 0.79

DeepSeek-V4-Pro

deepseek-ai

Permissive

MIT

Downloads: 610.6k
Likes: 5.6k
Updated: 2 months ago

View on HF →score 0.79

Gli strumenti che li fanno girare

Runtime, interfacce grafiche e server di inferenza, con i loro veri compromessi.

Tutti gli strumenti →

Ollama

Runtime

MIT

The fastest way to get a local LLM running with one command.

macOSLinuxWindows

Strengths

One-line install, one-line model pulls
Built-in OpenAI-compatible API on localhost:11434
Active model library with 4,500+ tagged variants

Trade-offs

Less raw throughput than vLLM under heavy concurrent load
Configuration is opinionated; advanced tuning means dropping into llama.cpp anyway

Visit project →Read install guide

llama.cpp

Runtime

MIT

Maximum control and the broadest hardware coverage in the open ecosystem.

macOSLinuxWindowsAndroidiOS

Strengths

Runs almost anywhere: CUDA, ROCm, Metal, Vulkan, CPU-only
Tight GGUF quantization control
Reference implementation behind most desktop LLM tools

Trade-offs

Command-line first; the UX assumes you read READMEs
Quantization options multiply quickly, easy to pick the wrong one

Visit project →Read install guide

DwarfStar (ds4)

Runtime

MIT

Running DeepSeek V4 Flash at usable speed on one 128 GB machine, and deliberately little else.

macOSLinux

Strengths

Asymmetric 2-bit quantization of the routed MoE experts only, which is what puts a 300B-class model inside 128 GB
Metal, CUDA and ROCm backends, with benchmarks published per machine rather than as a single headline number
On-disk KV cache, so a long session survives a restart

Trade-offs

Wants 96-128 GB of RAM to hold the Q2 build in memory, 256 GB or more for Q4; below that you are on the slower SSD-streaming path
Runs one model family on purpose: DeepSeek V4 Flash and PRO, plus GLM 5.2
The author calls it beta quality and the code changes fast

Visit project →No guide yet

LM Studio

GUI

Proprietary

Browsing, comparing and chatting with local models in a desktop GUI.

macOSLinuxWindows

Strengths

Polished chat UI with side-by-side model comparison
Built-in Hugging Face model browser
Local OpenAI-compatible API server with one click

Trade-offs

Closed source; the engine is llama.cpp but the shell is not
Less scriptable than CLI-first tools

Visit project →Read install guide

Perché prendersi il disturbo di eseguire l'AI in locale?

I grandi servizi cloud sono più facili per iniziare. Ma ci sono ragioni concrete per fare da soli. Eccone tre.

I tuoi dati restano tuoi

Quello che scrivi e quello che il modello risponde non lascia mai il tuo computer. Utile quando lavori con appunti personali, documenti di clienti, codice interno, o qualsiasi cosa che non incolleresti in un sito pubblico.

Funziona anche quando il cloud non funziona

Il file del modello vive sul tuo disco. Se l'azienda che lo ha creato chiude, alza i prezzi o cambia i termini, il tuo setup continua a funzionare. Il modello che scarichi oggi girerà ancora nel 2030, se il tuo computer lo farà.

Nessuna bolletta a sorpresa

L'AI in cloud si paga a consumo. L'AI locale costa il prezzo del tuo computer, più l'elettricità. Dopo il primo mese, il costo marginale di una domanda in più è praticamente zero.