What are the best local AI models for 32 GB?
The reviewed models that run well on 32 GB, ranked by each quality source.
Short answer
35models
35 reviewed text models run well on 32 GB, on a graphics card or a Mac. The strongest by each source:
- LMArena: Gemma 4 31B, Gemma 4 26B A4B, and Qwen3.8 27B
- LiveBench: Smaug Mini, Qwen3.8 27B, and Qwen3.6 27B
- Two separate rankings: the site never mixes them.
Computed from the reviewed catalog with the site's fit rule; scores from LiveBench and LMArena. Methodology
How this list is made
This page is computed from the site's catalog every time it is built, so it moves with the data instead of going stale. It takes every reviewed open text model with a published parameter count and asks the same question the hardware pages ask: does its Q4 file, plus working room and the machine's reserve, fit in fast memory? A model is listed here when it runs well on both reference machines: RTX 5090 · 32 GB VRAM, which keeps 2 GB of the card aside, and Mac mini M4 · 32 GB unified, which keeps 25% of its memory (8 GB) for macOS and your apps.
Quality comes from two sources that are never mixed: LiveBench (2026-06-25), test questions scored 0 to 100, and LMArena (2026-10-22), a rating from blind human votes. Each has its own table below. 18 of the 35 models have neither score yet; they still run well and are in the full table below. How to read the scores.
Top by LMArena
Blind human preference, 2026-10-22 release. Q4 need includes each machine's reserve.
| Model · publisher · parameters | LMArena 2026-10-22 (rating) | Q4 need, card / Mac (GB) |
|---|---|---|
| Gemma 4 31B Google · 30.7B | 1,452 | 26 / 32 |
| Gemma 4 26B A4B Google · 25.2B | 1,437 | 21 / 27 |
| Qwen3.8 27B Alibaba · 27B | 1,437 | 23 / 29 |
| Qwen3 30B A3B Instruct 2507 Alibaba · 30.5B | 1,382 | 25 / 31 |
| GLM-4.7 Flash Z.ai · 30B | 1,365 | 25 / 31 |
Top by LiveBench
Test questions, 2026-06-25 release, 0–100. Q4 need includes each machine's reserve.
| Model · publisher · parameters | LiveBench 2026-06-25 (0–100) | Q4 need, card / Mac (GB) |
|---|---|---|
| Smaug Mini Abacus AI · 27B | 76.9 | 23 / 29 |
| Qwen3.8 27B Alibaba · 27B | 75.3 | 23 / 29 |
| Qwen3.6 27B Alibaba · 27B | 64 | 23 / 29 |
Every model that runs well
35 reviewed text models, scored or not. Ordered by Hub downloads over 30 days, read 2026-10-25, not by quality; sort a score column to rank by that source alone.
| Qwen3 14B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | 14 / 20 | 9.0 | Apache 2.0 | 10.4M |
|---|---|---|---|---|---|---|
| Gemma 4 31B | Not in LiveBench 2026-06-25 | 1,452 | 26 / 32 | 18.3 | Apache 2.0 | 8.4M |
| Gemma 4 26B A4B | Not in LiveBench 2026-06-25 | 1,437 | 21 / 27 | 16.9 | Apache 2.0 | 8.2M |
| Qwen3 8B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | 9 / 15 | 5.0 | Apache 2.0 | 8.1M |
| Qwen3.8 27B | 75.3 | 1,437 | 23 / 29 | 16.5 | Apache 2.0 | 7.8M |
| Qwen3.5 9B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | 9 / 15 | 5.7 | Apache 2.0 | 6.1M |
| gpt-oss-20b | Not in LiveBench 2026-06-25 | 1,317 | 18 / 24 | 11.6 | Apache 2.0 | 5.8M |
| Llama 3.1 8B Instruct | Not in LiveBench 2026-06-25 | 1,211 | 8 / 14 | 4.9 | Llama 3.1 | 4.8M |
| Qwen3 1.7B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | 4 / 10 | 1.1 | Apache 2.0 | 4.1M |
| Gemma 4 E4B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | 8 / 14 | 5.0 | Apache 2.0 | 4M |
| Llama 3.2 1B Instruct | Not in LiveBench 2026-06-25 | 1,111 | 3 / 9 | 0.8 | Llama 3.2 | 3.5M |
| Gemma 4 E2B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | 6 / 12 | 3.1 | Apache 2.0 | 3M |
| Qwen3 4B Instruct 2507 | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | 5 / 11 | 2.5 | Apache 2.0 | 2.6M |
| Qwen3.6 27B | 64.0 | Not on the LMArena text leaderboard of 2026-10-22 | 23 / 29 | 16.8 | Apache 2.0 | 2.4M |
| Gemma 4 12B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | 11 / 17 | 7.1 | Apache 2.0 | 1.7M |
| Llama 3.2 3B Instruct | Not in LiveBench 2026-06-25 | 1,167 | 5 / 11 | 2.0 | Llama 3.2 | 1.1M |
| Qwen3 30B A3B Instruct 2507 | Not in LiveBench 2026-06-25 | 1,382 | 25 / 31 | 18.6 | Apache 2.0 | 1.1M |
| GLM-4.7 Flash | Not in LiveBench 2026-06-25 | 1,365 | 25 / 31 | 18.3 | MIT | 953.7K |
| Nemotron 3.5 Lightning 30B A3B | Not in LiveBench 2026-06-25 | Arena rated the NVFP4 build, NVIDIA-Nemotron-3.5-Lightning-30B-A3B (NVFP4); this entry is the BF16 reference, which NVIDIA reports with different scores | 25 / 31 | 18.9 | OpenMDW 1.1 | 861.5K |
| Qwen3 Coder 30B A3B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | 25 / 31 | 18.6 | Apache 2.0 | 745K |
| SmolLM3 3B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | 5 / 11 | 1.9 | Apache 2.0 | 718.7K |
| Phi-4 | Not in LiveBench 2026-06-25 | 1,256 | 13 / 19 | 8.0 | MIT | 368.3K |
| DeepSeek R1 0528 Qwen3 8B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | 9 / 15 | 5.0 | MIT | 355.8K |
| Phi-4 mini | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | 5 / 11 | 2.5 | MIT | 342.8K |
| Olmo 3 7B Instruct | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | 8 / 14 | 4.5 | Apache 2.0 | 262.5K |
| Devstral Small 2 24B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | 20 / 26 | 14.3 | Apache 2.0 | 195.6K |
| Mistral Small 3.2 24B | Not in LiveBench 2026-06-25 | 1,356 | 21 / 27 | 14.3 | Apache 2.0 | 181.4K |
| Ministral 3 14B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | 13 / 19 | 8.2 | Apache 2.0 | 126.5K |
| Ministral 3 3B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | 5 / 11 | 2.1 | Apache 2.0 | 87.4K |
| Ministral 3 8B | Not in LiveBench 2026-06-25 | Not on the LMArena text leaderboard of 2026-10-22 | 9 / 15 | 5.2 | Apache 2.0 | 79.4K |
| Granite 4.2 8B | Not in LiveBench 2026-06-25 | 1,290 | 8 / 14 | 5.3 | Apache 2.0 | 78.3K |
| Granite 4.2 3B | Not in LiveBench 2026-06-25 | 1,290 | 5 / 11 | 2.2 | Apache 2.0 | 39.8K |
| Granite 4.2 30B | Not in LiveBench 2026-06-25 | 1,340 | 25 / 31 | 17.7 | Apache 2.0 | 39.3K |
| Olmo 3 32B Think | Not in LiveBench 2026-06-25 | 1,306 | 26 / 32 | 19.5 | Apache 2.0 | 13.1K |
| Smaug Mini | 76.9 | Not on the LMArena text leaderboard of 2026-10-22 | 23 / 29 | 16.5 | Apache 2.0 | 817 |
The full lists
Each hardware page lists every model, scored or not, with files and sources.
- RTX 5090 · 32 GB VRAM
- 37 reviewed text models run well0 more run slowly with part of the model in PC RAM (32 GB assumed).
- Mac mini M4 · 32 GB unified
- 35 reviewed text models run wellThe Mac shares this memory with macOS and keeps 8 GB aside (the card keeps 2 GB), so its planning need is higher for the same file.
- Less memory
- What are the best local AI models for 16 GB?