Ruggi.SiteFree · in your browser · any OS
AI model comparison chart
Open-weight models you can run yourself and cloud tiers in one sortable table: size, context, licence, the memory a 4-bit copy needs, and per-token prices. Facts only, no rankings.
FreeNo signupWindows · macOS · Linux
The memory slider only filters open-weight rows (by their Q4_K_M memory figure); cloud rows need no memory of yours. Slide fully right to turn it off. Click a column title to sort; click again to reverse.
| Model | Type | Family / vendor | Size (B params) | Active (B) | Context (K tokens) | Licence | Min memory, Q4_K_M (GB) | Price in, $/1M | Price out, $/1M |
|---|
What the columns mean
- Size / Active: total parameters, in billions. Mixture-of-experts (MoE) models hold all of them in memory but use only the "active" ones for each word, so they run faster than their size suggests.
- Context: the largest context window listed on the public model card, in thousands of tokens. Cloud rows are blank because the list does not record it.
- Min memory, Q4_K_M: parameters × 4.85 ÷ 8 + 1 GB. 4.85 is the approximate bits per weight of the common 4-bit "Q4_K_M" file format, and the extra 1 GB is a rough allowance for the program and a short chat. Long contexts need more (see the "Can my PC run this AI model?" tool). Treat it as a minimum, not a promise.
- Price in / out: US dollars per one million input / output tokens for cloud tiers. Open-weight rows have no per-token price because you run them yourself (you pay for hardware and electricity).
- There are deliberately no "strengths" or "quality" columns: those are subjective and change with every release. Check independent benchmarks for the tasks you care about, then test with your own material.
Data last checked .