RunLocal

Model directory

AI models you can download and run yourself.

Each entry below is a family of AI models you can download for free and run on your own computer. The label on each card tells you who made it, how big it is, what licence it comes under, and what it is good at. New to all this? The glossary defines every term used here, or try the picker to find one that fits your machine.

China3 models

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

Current open-weight Qwen generation for local use. Qwen3.6-27B is the key workstation-class checkpoint. Qwen3.8-Max-Preview is newer but cloud/API-only as of August 15, 2026 and is intentionally excluded from the local picker.

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.

MiniCPM5-1B

OpenBMB · China

2026
License
Apache 2.0
Context
128k tokens
Sizes
1B
On-device inferenceEdge deploymentsLow-memory hardware

Edge-first model for phones and low-memory laptops.

United States3 models

Gemma 4

Google DeepMind · United States

2026
License
Apache 2.0
Context
128k–256k depending on checkpoint
Sizes
E2B · E4B · 12B · 26B-A4B MoE · 31B dense
On-device inferenceMultimodalConsumer GPUsWorkstations

Multimodal family spanning edge through workstation hardware. The 26B-A4B model activates about 4B parameters per token while retaining a 26B memory footprint.

Phi-4 family

Microsoft Research · United States

2025
License
MIT
Context
Varies by checkpoint
Sizes
Phi-4 Mini 3.8B · Phi-4 14B · reasoning variants
Edge devicesReasoning per parameterCost-sensitive inference

Verified Microsoft Phi family. RunLocal does not list an unverified Phi-5 family.

Llama 4 (Scout & Maverick)

Meta AI · United States

2025
License
Llama Community License (custom)
Context
Up to 10M tokens (Scout)
Sizes
Scout 109B MoE · Maverick 400B MoE
Long-context retrievalCodebase-scale RAGGeneral reasoning

Meta's open-weight family. Custom licensing applies; Scout and Maverick are MoE models with much smaller active parameter counts than total weights.

France (EU)2 models

Mistral Small 4

Mistral AI · France (EU)

2026
License
Apache 2.0
Context
256k tokens
Sizes
119B MoE (6.5B active)
ReasoningCoding agentsMultimodal workflowsEU-friendly deployments

Combines instruct, reasoning and coding modes in one multimodal MoE model. Low active parameters help compute speed, but the full weights still determine memory use.

Mistral Medium 3.5

Mistral AI · France (EU)

2026
License
Modified MIT / repository terms
Context
256k tokens
Sizes
128B dense
EU-friendly deploymentsCodingReasoningLong context

Dense 128B multimodal model. Local deployment is memory-heavy and generally workstation-class.

European Union1 model

EuroLLM-22B

EuroLLM Consortium · European Union

2025
License
Apache 2.0
Context
32k tokens
Sizes
1.7B · 9B · 22B
EU language coveragePublic-sector AIResearch

Transparent European family covering all 24 EU official languages plus additional languages.

United States (non-profit)1 model

OLMo 2

Allen Institute for AI · United States (non-profit)

2025
License
Apache 2.0
Context
8k tokens
Sizes
1B · 7B · 13B
Reproducible researchAuditable trainingEducation

Weights, data, training code and intermediate checkpoints are public; transparency is the differentiator.

Frontier open weights

The giants you (probably) can't run at home →

Kimi K3, GLM-5.2, Llama 4 Maverick, DeepSeek V4 Pro: what running them actually takes, and the runnable sibling from each family.

Trending · refreshed weekly

What the community is downloading this week →

The live top 16 from Hugging Face, ranked by downloads, likes and recency. Auto-updated every Monday.