128 GB · 90 models fit · data 2026-10-15

Best local LLM for 128 GB.

Qwen3.8 27B is the smartest model that fits in 128 GB. It needs about 22.7 GB and is about 59% as smart as Claude Opus 5.5.

Top picks for 128 GB, about 96 GB usable. AA = Artificial Analysis Intelligence Index; Claude Opus 5.5 = 57.6.
Model	Maker	Score (AA)	Needs 4-bit	Needs 8-bit	Run it
Qwen3.8 27B	Alibaba Qwen	33.7	22.7 GB	36.5 GB	`ollama run qwen3.8:27b`
K2 Horizon MoVA 36B-A4B	IFM (MBZUAI)	25.3	28.1 GB	47.8 GB	MLX only (Mac)
Qwen3.5 27B	Alibaba Qwen	22.9	21.5 GB	36.5 GB	`ollama run qwen3.5:27b`
Qwen3.6 27B	Alibaba Qwen	21.4	21.5 GB	36.5 GB	`ollama run qwen3.6:27b`
Nemotron 3.5 Bolt 74B-A9B	NVIDIA	21.4	30.7 GB	—	`ollama run nemotron-3.5-bolt:74b-a9b`
K2 Horizon 7B	IFM (MBZUAI)	20.6	8.6 GB	13 GB	MLX only (Mac)

By job

Pick by what you do.

CodingQwen3.8 27B

Needs 22.7 GB. Terminal-Bench 2.1: 79.8%.

ollama run qwen3.8:27b

Agents (uses tools)Qwen3.8 27B

Needs 22.7 GB.

ollama run qwen3.8:27b

Reads imagesQwen3.8 27B

Needs 22.7 GB.

ollama run qwen3.8:27b

New at 128 GB

What 128 GB adds over 96 GB.

Model	Needs	Score (AA)
Qwen3.5 122B-A10B	95.1 GB	17.7
Mistral Medium 3.5 128B	94 GB	14.2
gpt-oss-120b	74.2 GB	11.6
Mistral Small 4 119B	85 GB	11.3
GLM-4.5-Air 106B-A12B	85.9 GB	11.1

Machines

Computers and cards with 128 GB.

Graphics cards count VRAM, so they often fit a bit more.

Prices are approximate and dated. 2026 prices are inflated by the memory-chip shortage.
Machine	Type	Usable	Models that fit	Best fit	From (approx)
MacBook Pro 14"/16" (M5 Max)	Mac laptop	96 GB	90	Qwen3.8 27B	$4,099 · 2026-10-15
Mac Studio (M5 Max)	Mac desktop	96 GB	90	Qwen3.8 27B	$2,499 · 2026-10-15
NVIDIA DGX Spark (128GB)	NVIDIA AI box	96 GB	90	Qwen3.8 27B	$6,950 · 2026-10-15
NVIDIA Jetson AGX Thor Developer Kit	Dev kit	96 GB	90	Qwen3.8 27B	$5,499 · 2026-10-15
Framework Desktop (Ryzen AI Max+ 395)	AMD mini PC	96 GB	90	Qwen3.8 27B	$1,959 · 2026-10-15
GMKtec EVO-X2 (Ryzen AI Max+ 395)	AMD mini PC	96 GB	90	Qwen3.8 27B	$2,200 · 2026-10-15
Minisforum MS-S1 MAX (Ryzen AI Max+ 395)	AMD mini PC	96 GB	90	Qwen3.8 27B	$2,599 · 2026-10-15
Bosgame M5 (Ryzen AI Max+ 395)	AMD mini PC	96 GB	90	Qwen3.8 27B	$3,599 · 2026-10-15
Beelink GTR9 Pro (Ryzen AI Max+ 395)	AMD mini PC	96 GB	90	Qwen3.8 27B	$4,349 · 2026-10-15
ASUS ProArt GR1X (NVIDIA RTX Spark)	NVIDIA AI box	96 GB	90	Qwen3.8 27B	unknown
iBUYPOWER Pro AMD AI Halo Mini PC (Ryzen AI Max+ PRO 495)	AMD mini PC	96 GB	90	Qwen3.8 27B	$3,999 · 2026-10-16
Microsoft Surface RTX Spark Dev Box (128GB)	NVIDIA AI box	96 GB	90	Qwen3.8 27B	$6,000 · 2026-10-16
Sipeed MaixStation	Dev kit	96 GB	90	Qwen3.8 27B	unknown

Questions

Short answers.

What is the best local LLM for 128 GB of RAM?: Qwen3.8 27B by Alibaba Qwen. It scores 33.7 on the Artificial Analysis Intelligence Index and needs about 22.7 GB at 4-bit. That is about 59% of Claude Opus 5.5 (57.6), the top paid model. Data as of 2026-10-15.
How many local models fit in 128 GB?: 90 of the 124 text models in our data fit. We assume about 96 GB is usable for the model (75% of 128 GB), and a model needs its 4-bit file size × 1.15 + 2 GB.
What is the biggest model that runs in 128 GB?: Mistral Medium 3.5 128B, 127.7B parameters, needs about 94 GB at 4-bit. Bigger is not always smarter: Qwen3.8 27B scores 33.7; Mistral Medium 3.5 128B scores 14.2.
What is the best coding model for 128 GB?: Qwen3.8 27B (needs 22.7 GB, Terminal-Bench 2.1 79.8%). Run it in a coding app like OpenCode or Cline. See https://insidethemachine.ai/best/local-coding-model.
Does 128 GB of VRAM work the same as 128 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 128 GB card fits a little more than 128 GB of Mac memory. Models that spill out of VRAM into normal RAM run much slower.

Scores are from Artificial Analysis and LMArena, not our own tests yet. Memory need = 4-bit file × 1.15 + 2 GB; a model fits when that is at most 75% of memory. Models 2026-10-15, machines 2026-10-15.

AGENT VIEW · text/markdown · or send Accept: text/markdown

# Best local LLM for 128GB RAM or VRAM

URL: https://insidethemachine.ai/best/local-llm-for-128gb-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: 96 GB (75% of 128). 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: 90 of 124.

## Top picks

| model | maker | AA index | needs (4-bit) | needs (8-bit) | run |
|---|---|---|---|---|---|
| Qwen3.8 27B | Alibaba Qwen | 33.7 | 22.7 GB | 36.5 GB | `ollama run qwen3.8:27b` |
| 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` |
| Qwen3.6 27B | Alibaba Qwen | 21.4 | 21.5 GB | 36.5 GB | `ollama run qwen3.6:27b` |
| Nemotron 3.5 Bolt 74B-A9B | NVIDIA | 21.4 | 30.7 GB | - | `ollama run nemotron-3.5-bolt:74b-a9b` |
| K2 Horizon 7B | IFM (MBZUAI) | 20.6 | 8.6 GB | 13 GB | MLX only |

## By job

- Coding: Qwen3.8 27B (22.7 GB, Terminal-Bench 2.1 79.8%) `ollama run qwen3.8:27b`
- Agents (tool use): Qwen3.8 27B (22.7 GB) `ollama run qwen3.8:27b`
- Vision (reads images): Qwen3.8 27B (22.7 GB)

## New at 128 GB (did not fit in 96 GB)

- Qwen3.5 122B-A10B (95.1 GB, AA 17.7)
- Mistral Medium 3.5 128B (94 GB, AA 14.2)
- gpt-oss-120b (74.2 GB, AA 11.6)
- Mistral Small 4 119B (85 GB, AA 11.3)
- GLM-4.5-Air 106B-A12B (85.9 GB, AA 11.1)

## Machines with 128 GB

| machine | type | usable GB | models that fit | best fit | from (approx) |
|---|---|---|---|---|---|
| MacBook Pro 14"/16" (M5 Max) | Mac laptop | 96 | 90 | Qwen3.8 27B | $4,099 (2026-10-15) |
| Mac Studio (M5 Max) | Mac desktop | 96 | 90 | Qwen3.8 27B | $2,499 (2026-10-15) |
| NVIDIA DGX Spark (128GB) | NVIDIA AI box | 96 | 90 | Qwen3.8 27B | $6,950 (2026-10-15) |
| NVIDIA Jetson AGX Thor Developer Kit | Dev kit | 96 | 90 | Qwen3.8 27B | $5,499 (2026-10-15) |
| Framework Desktop (Ryzen AI Max+ 395) | AMD mini PC | 96 | 90 | Qwen3.8 27B | $1,959 (2026-10-15) |
| GMKtec EVO-X2 (Ryzen AI Max+ 395) | AMD mini PC | 96 | 90 | Qwen3.8 27B | $2,200 (2026-10-15) |
| Minisforum MS-S1 MAX (Ryzen AI Max+ 395) | AMD mini PC | 96 | 90 | Qwen3.8 27B | $2,599 (2026-10-15) |
| Bosgame M5 (Ryzen AI Max+ 395) | AMD mini PC | 96 | 90 | Qwen3.8 27B | $3,599 (2026-10-15) |
| Beelink GTR9 Pro (Ryzen AI Max+ 395) | AMD mini PC | 96 | 90 | Qwen3.8 27B | $4,349 (2026-10-15) |
| ASUS ProArt GR1X (NVIDIA RTX Spark) | NVIDIA AI box | 96 | 90 | Qwen3.8 27B | unknown |
| iBUYPOWER Pro AMD AI Halo Mini PC (Ryzen AI Max+ PRO 495) | AMD mini PC | 96 | 90 | Qwen3.8 27B | $3,999 (2026-10-16) |
| Microsoft Surface RTX Spark Dev Box (128GB) | NVIDIA AI box | 96 | 90 | Qwen3.8 27B | $6,000 (2026-10-16) |
| Sipeed MaixStation | Dev kit | 96 | 90 | Qwen3.8 27B | unknown |

## Questions

**What is the best local LLM for 128 GB of RAM?**
Qwen3.8 27B by Alibaba Qwen. It scores 33.7 on the Artificial Analysis Intelligence Index and needs about 22.7 GB at 4-bit. That is about 59% of Claude Opus 5.5 (57.6), the top paid model. Data as of 2026-10-15.

**How many local models fit in 128 GB?**
90 of the 124 text models in our data fit. We assume about 96 GB is usable for the model (75% of 128 GB), and a model needs its 4-bit file size × 1.15 + 2 GB.

**What is the biggest model that runs in 128 GB?**
Mistral Medium 3.5 128B, 127.7B parameters, needs about 94 GB at 4-bit. Bigger is not always smarter: Qwen3.8 27B scores 33.7; Mistral Medium 3.5 128B scores 14.2.

**What is the best coding model for 128 GB?**
Qwen3.8 27B (needs 22.7 GB, Terminal-Bench 2.1 79.8%). Run it in a coding app like OpenCode or Cline. See https://insidethemachine.ai/best/local-coding-model.

**Does 128 GB of VRAM work the same as 128 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 128 GB card fits a little more than 128 GB of Mac memory. Models that spill out of VRAM into normal RAM run much slower.

More: https://insidethemachine.ai/models?tier=128 · https://insidethemachine.ai/machines?tier=128 · https://insidethemachine.ai/best