# Best local LLM for 256GB RAM or VRAM

URL: https://insidethemachine.ai/best/local-llm-for-256gb-ram
Data: models 2026-10-15, machines 2026-10-15. Source of scores: Artificial Analysis Intelligence Index (AA), LMArena. Not our own tests yet.
Usable memory assumed: 192 GB (75% of 256). Rule: min_memory_gb = weights_gb_4bit × 1.15 + 2. weights_gb_4bit = Q4_K_M file (Ollama, else HF GGUF) → MLX 4-bit (HF) → Ollama MLX/NVFP4 tag → Ollama default tag → params_total_b × 0.6 (flagged weights_estimated). +15% ≈ KV cache at ~8–16K context; +2GB runtime/process overhead. Long contexts (128K+) need much more.
Models that fit: 95 of 124.

## Top picks

| model | maker | AA index | needs (4-bit) | needs (8-bit) | run |
|---|---|---|---|---|---|
| Qwen3.8-Flash-Next (125B-A6B) | Alibaba Qwen | 39.8 | 140 GB | 219.3 GB | `ollama run qwen3.8-flash-next:125b-a6b` |
| Qwen3.8 27B | Alibaba Qwen | 33.7 | 22.7 GB | 36.5 GB | `ollama run qwen3.8:27b` |
| Nemotron 3.5 Super VL 120B-A12B | NVIDIA | 26.5 | 105.5 GB | 154.9 GB | `ollama run nemotron-3.5-super:120b-a12b` |
| K2 Horizon MoVA 36B-A4B | IFM (MBZUAI) | 25.3 | 28.1 GB | 47.8 GB | MLX only |
| Qwen3.5 27B | Alibaba Qwen | 22.9 | 21.5 GB | 36.5 GB | `ollama run qwen3.5:27b` |
| MiniMax M2.7 | MiniMax | 22.8 | 161.1 GB | 281.6 GB | MLX only |

## By job

- Coding: Qwen3.8-Flash-Next (125B-A6B) (140 GB, Terminal-Bench 2.1 86.1%) `ollama run qwen3.8-flash-next:125b-a6b`
- Agents (tool use): Qwen3.8-Flash-Next (125B-A6B) (140 GB) `ollama run qwen3.8-flash-next:125b-a6b`
- Vision (reads images): Qwen3.8-Flash-Next (125B-A6B) (140 GB)

## New at 256 GB (did not fit in 192 GB)

- MiniMax M2.7 (161.1 GB, AA 22.8)

## Machines with 256 GB

| machine | type | usable GB | models that fit | best fit | from (approx) |
|---|---|---|---|---|---|
| Mac Studio (M5 Ultra) | Mac desktop | 192 | 95 | Qwen3.8-Flash-Next (125B-A6B) | $5,499 (2026-10-15) |

## Questions

**What is the best local LLM for 256 GB of RAM?**
Qwen3.8-Flash-Next (125B-A6B) by Alibaba Qwen. It scores 39.8 on the Artificial Analysis Intelligence Index and needs about 140 GB at 4-bit. That is about 69% of Claude Opus 5.5 (57.6), the top paid model. Data as of 2026-10-15.

**How many local models fit in 256 GB?**
95 of the 124 text models in our data fit. We assume about 192 GB is usable for the model (75% of 256 GB), and a model needs its 4-bit file size × 1.15 + 2 GB.

**What is the biggest model that runs in 256 GB?**
MiniMax M2.7, 228.7B parameters, needs about 161.1 GB at 4-bit. Bigger is not always smarter: Qwen3.8-Flash-Next (125B-A6B) scores 39.8; MiniMax M2.7 scores 22.8.

**What is the best coding model for 256 GB?**
Qwen3.8-Flash-Next (125B-A6B) (needs 140 GB, Terminal-Bench 2.1 86.1%). Run it in a coding app like OpenCode or Cline. See https://insidethemachine.ai/best/local-coding-model.

**Does 256 GB of VRAM work the same as 256 GB of RAM?**
Close. On a Mac the memory is shared, and about 75% goes to the model. On a PC the graphics card's VRAM is what counts; we estimate VRAM minus 1 GB is usable, so a 256 GB card fits a little more than 256 GB of Mac memory. Models that spill out of VRAM into normal RAM run much slower.

More: https://insidethemachine.ai/models?tier=256 · https://insidethemachine.ai/machines?tier=256 · https://insidethemachine.ai/best