Best local LLM in 2026: rankings by use case, VRAM, and machine
Ranking updated on 09/10/2026
72 thematic rankings to find the open-source LLM best suited to your needs, hardware, or legal constraints. All listed models are genuinely self-hostable (≤ 100B parameters) — no cloud LLMs or models inaccessible locally. The scoring is transparent: view the methodology.
By use case
The most practical criterion—your everyday needs.
By hardware
Your GPU or Mac determines what can run locally.
By constraint
License and origin — often the real filter for businesses.
How we rank them
Each ranking applies a filter (tags, size, license, VRAM), then a weighted score that favors credible models for mainstream local use: sweet-spot size of 7–32B, permissive license, recent release, and specialization where relevant.
Models beyond 100B parameters are excluded by default — they require hardware inaccessible to 99% of visitors. Exception for the list Sovereign LLM which goes up to 200B to include Mistral Large as a reference.
Need an unlisted combination? The configurator evaluates every model in the catalog on your exact hardware.