Modelli AI che puoi scaricare ed eseguire tu stesso.

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

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.

License

MIT

Context

1M tokens (65k native, extended 16x with YaRN)

Sizes

304B-A13B MoE (284B before the DSpark decoder)

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.

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

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.

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

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

License

MIT

Context

202k tokens

Sizes

31B MoE

The runnable member of the GLM family, and the counterpart to the frontier-scale GLM-5.2: MIT licensed, 200k context, mature GGUF ecosystem, and one of the strongest agentic-coding models that fits on a single workstation GPU.

License

Apache 2.0

Context

128k tokens

Sizes

1B

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

License

Apache 2.0

Context

128k–256k depending on checkpoint

Sizes

E2B · E4B · 12B · 26B-A4B MoE · 31B dense

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

License

MIT

Context

Varies by checkpoint

Sizes

Phi-4 Mini 3.8B · Phi-4 14B · reasoning variants

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

License

Llama Community License (custom)

Context

Up to 10M tokens (Scout)

Sizes

Scout 109B MoE · Maverick 400B MoE

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

License

Apache 2.0

Context

131k tokens

Sizes

20B MoE (MXFP4 weights) · 120B MoE

OpenAI's open-weight release and, by download count, the most used open model of the past year. Ships natively in MXFP4, so the 20B fits a 16 GB card at the precision it was trained for. Apache 2.0, no usage caps, no licence acceptance step.

License

NVIDIA Open Model License (custom — read before commercial use)

Context

262k tokens

Sizes

30B-A3B hybrid Mamba/MoE (~3B active)

August 2026. A hybrid Mamba-attention MoE: 31B of weights, roughly 3B active per token, so it answers at small-model speed. GGUF builds come from ggml-org and unsloth. The licence is custom, not OSI-approved.

License

Custom Liquid AI licence (check the model card)

Context

131k tokens

Sizes

2.6B

July 2026. A convolution-attention hybrid built for edge hardware, covering 16 languages at 2.6B parameters. Liquid publishes its own GGUF files, which is why it reached the llama.cpp crowd faster than most small models.

License

Apache 2.0

Context

256k tokens

Sizes

119B MoE (6.5B active)

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.

License

Modified MIT / repository terms

Context

256k tokens

Sizes

128B dense

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

License

Apache 2.0

Context

32k tokens

Sizes

1.7B · 9B · 22B

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

License

Apache 2.0

Context

65k tokens

Sizes

7B · 32B (Instruct and Think variants)

Weights, data, training code and intermediate checkpoints are public; transparency is the differentiator. Olmo 3.1 (December 2025) added reasoning-tuned Think variants at 7B and 32B, so the transparency argument costs far less capability than it did with OLMo 2.