Twoody
Local LLM calculator
The best local LLM for your RAM
Memory decides which language models fit your computer. For each size, from 8 to 512 GB: the strongest models that run comfortably on a Mac with Apple silicon and on a PC without a graphics card, how fast they write, and the model Twoody suggests by default.
| Memory | Computer | Strongest that runs comfortably | Writes | Suggested by Twoody by default |
|---|---|---|---|---|
8 GB of memory |
Mac with Apple silicon A18 Pro — MacBook Neo | Qwen3.5 4B One click in Twoody Also: Qwen3 4B Instruct 2507, Nemotron 3 Nano 4B | 9.9–18 tokens/s | Qwen3.5 4B (4 GB) |
| PC without a graphics card DDR5 memory | Qwen3.5 0.8B One click in Twoody Also: LFM2.5 2.6B, Llama 3.2 1B | 34–103 tokens/s | Qwen3.5 4B (4 GB) Runs — close other apps | |
16 GB of memory |
Mac with Apple silicon M6 — Mac mini | Gemma 4 12B One click in Twoody Also: GLM-4.6V Flash, Qwen3.5 9B | 12–22 tokens/s | Qwen3.5 9B (7.4 GB) |
| PC without a graphics card DDR5 memory | GLM-4.6V Flash One click in Twoody Also: Qwen3.5 9B, GLM-4 9B | 5.4–9.7 tokens/s | Qwen3.5 4B (4.4 GB) | |
24 GB of memory |
Mac with Apple silicon M5 Pro — MacBook Pro, Mac mini | Qwen3 14B One click in Twoody Also: Phi-4 14B, Ministral 3 14B | 22–32 tokens/s | Qwen3.5 9B (7.4 GB) |
32 GB of memory |
Mac with Apple silicon M6 — Mac mini | Qwen3.6 35B-A3B One click in Twoody Also: Gemma 4 31B, Qwen3 30B-A3B Instruct 2507 | 41–73 tokens/s | Qwen3.5 9B (7.9 GB) |
| PC without a graphics card DDR5 memory | Qwen3.6 35B-A3B One click in Twoody Also: Qwen3.6 35B-A3B (4 bits), Qwen3 30B-A3B Instruct 2507 | 18–31 tokens/s | Qwen3.5 9B (7.9 GB) | |
48 GB of memory |
Mac with Apple silicon M5 Max — MacBook Pro, Mac Studio | Qwen3.6 35B-A3B One click in Twoody Also: Qwen3.6 35B-A3B (4 bits), Qwen3.5 35B-A3B | 100+ tokens/s | Qwen3.5 9B (7.9 GB) |
64 GB of memory |
Mac with Apple silicon M5 Max — MacBook Pro, Mac Studio | Qwen3-Next 80B-A3B Not in Twoody's catalogue Also: Llama 3.3 70B, Qwen3.6 35B-A3B | 98–196 tokens/s | Qwen3.5 9B (7.9 GB) |
| PC without a graphics card DDR5 memory | Qwen3-Next 80B-A3B Not in Twoody's catalogue Also: Qwen3.6 35B-A3B, Qwen3.6 35B-A3B (4 bits) | 13–23 tokens/s | Qwen3.5 9B (7.9 GB) | |
128 GB of memory |
Mac with Apple silicon M5 Max — MacBook Pro, Mac Studio | Qwen3.5 122B-A10B Not in Twoody's catalogue Also: Mistral Small 4 119B*, Qwen3-Next 80B-A3B* | 36–73 tokens/s | Qwen3.5 9B (7.9 GB) |
| PC without a graphics card DDR5 memory | Mistral Small 4 119B Not in Twoody's catalogue Also: Qwen3-Next 80B-A3B*, Qwen3.6 35B-A3B | 6.7–12 tokens/s | Qwen3.5 9B (7.9 GB) | |
256 GB of memory |
Mac with Apple silicon M5 Ultra — Mac Studio | Qwen3 235B-A22B Instruct 2507 Not in Twoody's catalogue Also: Qwen3.5 122B-A10B*, Mistral Small 4 119B* | 28–50 tokens/s | Qwen3.5 9B (7.9 GB) |
512 GB of memory |
Mac with Apple silicon M5 Ultra — Mac Studio | Qwen3 235B-A22B Instruct 2507 Not in Twoody's catalogue Also: Qwen3.5 122B-A10B*, Mistral Small 4 119B* | 28–50 tokens/s | Qwen3.5 9B (7.9 GB) |
* Not in Twoody's catalogue; the others install in one click.
“Strongest that runs comfortably”: the strongest model that fits comfortably in this memory and writes there at least at your reading speed (5 tokens per second). The Mac line is the first Mac with Apple silicon that comes with this memory, its chip named below; the PC line, a PC whose processor runs the model, with DDR5 memory. “Suggested by Twoody by default”: the model the app installs first there, and the memory it needs.
Speeds are estimates, not measurements made with Twoody, computed from public llama.cpp benchmarks. Figures reviewed on October 16, 2026.
On a Mac
A Mac shares its memory between the processor and the graphics, which may use most of it: the more memory, the larger the models it runs, and that memory is read fast.
On a PC without a graphics card
The model runs on the processor, which reads the memory more slowly: with the same memory, a PC writes more slowly than a Mac, and the strongest model that keeps up with your reading is smaller. DDR5 memory writes faster than DDR4; integrated graphics change the speeds, and the calculator gives them for your processor.
With a graphics card
On a PC with a graphics card, the card's own memory (VRAM) decides: a model that fits it writes much faster than on the processor, and one that does not slows down a lot. The pages on NVIDIA and AMD cards give, card by card, the strongest model that fits.
Questions about running an LLM locally
What is the best local LLM for 8 GB of RAM?
With 8 GB, the strongest model that runs comfortably is Qwen3.5 4B on a Mac with Apple silicon (9.9–18 tokens/s with the A18 Pro chip) and Qwen3.5 0.8B on a PC without a graphics card, with DDR5 memory (34–103 tokens/s). Both install in one click from Twoody's catalogue. Twoody suggests Qwen3.5 4B by default on both, which needs 4 GB of memory there: one of the models it has measured with its memory, its tools and your documents.
Which local LLM for 16 GB of RAM?
With 16 GB, the strongest model that runs comfortably is Gemma 4 12B on a Mac with Apple silicon (12–22 tokens/s with the M6 chip) and GLM-4.6V Flash on a PC without a graphics card, with DDR5 memory (5.4–9.7 tokens/s). Both install in one click from Twoody's catalogue. By default, Twoody suggests Qwen3.5 9B on that Mac (7.4 GB of memory) and Qwen3.5 4B on that PC (4.4 GB): models it has measured with its memory, its tools and your documents.
What can I run with 32 GB of RAM?
With 32 GB, the strongest model that runs comfortably is Qwen3.6 35B-A3B, on a Mac with Apple silicon as on a PC without a graphics card: 41–73 tokens/s on a Mac with the M6 chip, 18–31 tokens/s on a PC with DDR5 memory. It installs in one click from Twoody's catalogue. Twoody suggests Qwen3.5 9B by default on both, which needs 7.9 GB of memory there: one of the models it has measured with its memory, its tools and your documents.
Which LLM for 64 GB of memory?
With 64 GB, the strongest model that runs comfortably is Qwen3-Next 80B-A3B, on a Mac with Apple silicon as on a PC without a graphics card: 98–196 tokens/s on a Mac with the M5 Max chip, 13–23 tokens/s on a PC with DDR5 memory. It is not in Twoody's catalogue. Twoody suggests Qwen3.5 9B by default on both, which needs 7.9 GB of memory there: one of the models it has measured with its memory, its tools and your documents.
RAM or graphics card memory: which one counts?
On a Mac, one memory serves the processor and the graphics: that is the one that counts. On a PC with a graphics card, the card's own memory (VRAM) decides which models run fast; a model larger than it spills into the computer's memory and slows down a lot. Without a graphics card, the computer's memory decides, and the processor sets the speed.
Does more memory make a model faster?
No: memory decides which models fit, not how fast they write. Speed follows how fast the memory is read, its bandwidth: that is why a Pro, Max or Ultra chip, or a graphics card, writes faster with the same model.
The best local LLM, by computer
Try Twoody before it opens to everyone.
The private beta is free, on Mac, Windows and Linux, and on iPhone and Android with Twoody on your computer. Leave your email to ask for an invitation: we will write to you when it is ready for you.