01
Read the machine.
OS, RAM, and VRAM set the ceiling.
Local LLM fit test
Which model fits?
Rank local LLMs by your RAM, VRAM, and use case.
Catalog leaderboard
Top models by general score
| # | Model | General | Context | Downloads |
|---|---|---|---|---|
| 01 | Hy3 tencent / hy3 family | 92.9 | 262K | |
| 02 | Hy4-preview tencent / hy4 family | 92.3 | 1.0M | |
| 03 | step-3.5-flash AesSedai / step family | 89.8 | — | |
| 04 | minimax-m2.7 AaryanK / minimax family | 89.4 | — | |
| 05 | DeepSeek-R1 deepseek-ai / DeepSeek family | 87.3 | — |
How the budget is computed
Memory is the first filter.
02
Reserve the overhead.
2 GB reserved. macOS uses 75% of unified RAM.
03
Rank what loads.
Only variants at or below the budget are eligible.