Pick a model for my computer

Choose your computer and what you want to do. We pick the model with the highest independent quality score that fits, not the most popular one.

Scores are the Artificial Analysis Intelligence Index, an independent test of overall quality (higher is better). * means Artificial Analysis marks the score as an estimate. The index measures overall quality; we could not get separate coding or image scores, so the same score ranks every use. Our estimate from published download sizes. We did not test these ourselves. Fits = download x 1.25 + 2 GB fits in usable memory, which leaves room for longer chats. Tight = only download + 1.5 GB fits, so short chats work but long chats or documents may spill onto the CPU and slow down. Usable memory is the card's full VRAM, or on a Mac 2/3 of RAM up to 32 GB and 3/4 at 64 GB+ (macOS keeps part of memory away from the GPU by default; Apple says a 32 GB Mac gives the GPU 21 GB and a 64 GB Mac 48 GB). On an AMD Ryzen AI Max+ PC with 128 GB it is the graphics memory you set in AMD Software, up to 96 GB per AMD. Apple · AMD Not sure how much graphics memory you have? On a Mac, open Apple menu > About This Mac. On Windows, open Task Manager > Performance > GPU and look at "Dedicated GPU memory". On an AMD Ryzen AI Max+ PC, first set Variable Graphics Memory to 96 GB in AMD Software: Adrenalin Edition > Performance > Tuning, then restart (AMD's steps). Ollama lists the Ryzen AI Max+ 395 as a supported GPU (Ollama docs).

Before long chats or documents: Ollama's default context is only 4K tokens on cards under 24 GB, and 32K on 24-48 GB. Raise it with the slider in Ollama's Settings, or older parts of the chat get dropped without warning. source

8 GB graphics card · Everyday chat

Best pick
Tight LFM2.5-24B-A2B lfm2.5:8b · 5.2 GB download · AA index 7* (LFM2.5-8B-A1B)
Highest independent score among the models we track that run on this machine. It is tight here: short chats are fine, long chats may slow down. It writes reasoning text before each answer; that is normal.
```
ollama run lfm2.5:8b
```

Also fits, but has no independent score yet so we can't rank it: MiniCPM-V 4.7 (35B-A3B) (openbmb/minicpm-v4.7:1b).

8 GB graphics card · Coding help

Best pick
Tight LFM2.5-24B-A2B lfm2.5:8b · 5.2 GB download · AA index 7* (LFM2.5-8B-A1B)
Highest independent score among the models we track that run on this machine. It is tight here: short chats are fine, long chats may slow down. It writes reasoning text before each answer; that is normal.
```
ollama run lfm2.5:8b
```

8 GB graphics card · Reading images

No model with an independent score passes our fit rule for this yet. We would rather say so than suggest something that will crawl.

Also fits, but has no independent score yet so we can't rank it: MiniCPM-V 4.7 (35B-A3B) (openbmb/minicpm-v4.7:1b).

8 GB graphics card · Long documents

No model with an independent score passes our fit rule for this yet (for long documents we only pick models that Fit, not Tight). We would rather say so than suggest something that will crawl.

16 GB graphics card · Everyday chat

Best pick
Fits Gemma 4 (26B-A4B / 31B) gemma4:12b · 8 GB download · AA index 14* (12B)
Highest independent score among the models we track that fit this machine. Every Gemma 4 size reads images.
```
ollama run gemma4:12b
```

Also fits, but has no independent score yet so we can't rank it: MiniCPM-V 4.7 (35B-A3B) (openbmb/minicpm-v4.7:1b).

16 GB graphics card · Coding help

Best pick
Fits Gemma 4 (26B-A4B / 31B) gemma4:12b · 8 GB download · AA index 14* (12B)
Highest independent score among the models we track that fit this machine. Every Gemma 4 size reads images.
```
ollama run gemma4:12b
```

16 GB graphics card · Reading images

Best pick
Fits Gemma 4 (26B-A4B / 31B) gemma4:12b · 8 GB download · AA index 14* (12B)
Highest independent score among the models we track that fit this machine. Every Gemma 4 size reads images.
```
ollama run gemma4:12b
```

Also fits, but has no independent score yet so we can't rank it: MiniCPM-V 4.7 (35B-A3B) (openbmb/minicpm-v4.7:1b).

16 GB graphics card · Long documents

Best pick
Fits Gemma 4 (26B-A4B / 31B) gemma4:12b · 8 GB download · AA index 14* (12B)
Highest independent score among the models we track that fit this machine. Every Gemma 4 size reads images.
```
ollama run gemma4:12b
```

24 GB graphics card · Everyday chat

Best pick
Tight Qwen3.8-27B qwen3.8:27b · 18 GB download · AA index 28 (medium, Ollama's default thinking level)
Highest independent score among the models we track that run on this machine. It is tight here: short chats are fine, long chats may slow down. For quick answers, type /set nothink in the chat.
```
ollama run qwen3.8:27b
```
Alternative
Fits Gemma 4 (26B-A4B / 31B) gemma4:12b · 8 GB download · AA index 14* (12B)
Why consider it: Fits with room for long chats. Qwen3.8 is tight on this card, but scores higher. Every Gemma 4 size reads images.
```
ollama run gemma4:12b
```

Also fits, but has no independent score yet so we can't rank it: MiniCPM-V 4.7 (35B-A3B) (openbmb/minicpm-v4.7:1b), LFM2.5-24B-A2B (lfm2.5:24b), Laguna XS 2.1 (33B-A3B) (laguna-xs-2.1).

24 GB graphics card · Coding help

Best pick
Tight Qwen3.8-27B qwen3.8:27b · 18 GB download · AA index 28 (medium, Ollama's default thinking level)
Highest independent score among the models we track that run on this machine. It is tight here: short chats are fine, long chats may slow down. For quick answers, type /set nothink in the chat.
```
ollama run qwen3.8:27b
```
Alternative
Fits Gemma 4 (26B-A4B / 31B) gemma4:12b · 8 GB download · AA index 14* (12B)
Why consider it: Fits with room for long coding sessions. Qwen3.8 is tight on this card, but scores higher. Every Gemma 4 size reads images.
```
ollama run gemma4:12b
```

Also fits, but has no independent score yet so we can't rank it: LFM2.5-24B-A2B (lfm2.5:24b), Laguna XS 2.1 (33B-A3B) (laguna-xs-2.1).

24 GB graphics card · Reading images

Best pick
Tight Qwen3.8-27B qwen3.8:27b · 18 GB download · AA index 28 (medium, Ollama's default thinking level)
Highest independent score among the models we track that run on this machine. It is tight here: short chats are fine, long chats may slow down. For quick answers, type /set nothink in the chat.
```
ollama run qwen3.8:27b
```
Alternative
Fits Gemma 4 (26B-A4B / 31B) gemma4:12b · 8 GB download · AA index 14* (12B)
Why consider it: Fits with room to spare. Qwen3.8 is tight on this card, but scores higher. Every Gemma 4 size reads images.
```
ollama run gemma4:12b
```

Also fits, but has no independent score yet so we can't rank it: MiniCPM-V 4.7 (35B-A3B) (openbmb/minicpm-v4.7:1b).

24 GB graphics card · Long documents

Best pick
Fits Gemma 4 (26B-A4B / 31B) gemma4:12b · 8 GB download · AA index 14* (12B)
Highest independent score among the models we track that fit this machine. Every Gemma 4 size reads images.
```
ollama run gemma4:12b
```

16 GB Mac · Everyday chat

Best pick
Tight Gemma 4 (26B-A4B / 31B) gemma4:e2b · 7.5 GB download · AA index 8* (E2B)
Highest independent score among the models we track that run on this machine. It is tight here: short chats are fine, long chats may slow down. Every Gemma 4 size reads images.
```
ollama run gemma4:e2b
```

Also fits, but has no independent score yet so we can't rank it: MiniCPM-V 4.7 (35B-A3B) (openbmb/minicpm-v4.7:1b).

16 GB Mac · Coding help

Best pick
Tight Gemma 4 (26B-A4B / 31B) gemma4:e2b · 7.5 GB download · AA index 8* (E2B)
Highest independent score among the models we track that run on this machine. It is tight here: short chats are fine, long chats may slow down. Every Gemma 4 size reads images.
```
ollama run gemma4:e2b
```

16 GB Mac · Reading images

Best pick
Tight Gemma 4 (26B-A4B / 31B) gemma4:e2b · 7.5 GB download · AA index 8* (E2B)
Highest independent score among the models we track that run on this machine. It is tight here: short chats are fine, long chats may slow down. Every Gemma 4 size reads images.
```
ollama run gemma4:e2b
```

Also fits, but has no independent score yet so we can't rank it: MiniCPM-V 4.7 (35B-A3B) (openbmb/minicpm-v4.7:1b).

16 GB Mac · Long documents

No model with an independent score passes our fit rule for this yet (for long documents we only pick models that Fit, not Tight). We would rather say so than suggest something that will crawl.

24 GB Mac · Everyday chat

Best pick
Fits Gemma 4 (26B-A4B / 31B) gemma4:e4b · 9.5 GB download · AA index 9 (E4B)
Highest independent score among the models we track that fit this machine. Every Gemma 4 size reads images.
```
ollama run gemma4:e4b
```

Also fits, but has no independent score yet so we can't rank it: MiniCPM-V 4.7 (35B-A3B) (openbmb/minicpm-v4.7:1b).

24 GB Mac · Coding help

Best pick
Fits Gemma 4 (26B-A4B / 31B) gemma4:e4b · 9.5 GB download · AA index 9 (E4B)
Highest independent score among the models we track that fit this machine. Every Gemma 4 size reads images.
```
ollama run gemma4:e4b
```

24 GB Mac · Reading images

Best pick
Fits Gemma 4 (26B-A4B / 31B) gemma4:e4b · 9.5 GB download · AA index 9 (E4B)
Highest independent score among the models we track that fit this machine. Every Gemma 4 size reads images.
```
ollama run gemma4:e4b
```

Also fits, but has no independent score yet so we can't rank it: MiniCPM-V 4.7 (35B-A3B) (openbmb/minicpm-v4.7:1b).

24 GB Mac · Long documents

Best pick
Fits Gemma 4 (26B-A4B / 31B) gemma4:e4b · 9.5 GB download · AA index 9 (E4B)
Highest independent score among the models we track that fit this machine. Every Gemma 4 size reads images.
```
ollama run gemma4:e4b
```

32 GB Mac · Everyday chat

Best pick
Tight Qwen3.8-27B qwen3.8:27b · 18 GB download · AA index 28 (medium, Ollama's default thinking level)
Highest independent score among the models we track that run on this machine. It is tight here: short chats are fine, long chats may slow down. For quick answers, type /set nothink in the chat.
```
ollama run qwen3.8:27b
```
Alternative
Fits Gemma 4 (26B-A4B / 31B) gemma4:e4b · 9.5 GB download · AA index 9 (E4B)
Why consider it: Fits with room for long chats. Qwen3.8 is tight on a 32 GB Mac, but scores much higher. Every Gemma 4 size reads images.
```
ollama run gemma4:e4b
```

Also fits, but has no independent score yet so we can't rank it: MiniCPM-V 4.7 (35B-A3B) (openbmb/minicpm-v4.7:1b), LFM2.5-24B-A2B (lfm2.5:24b).

32 GB Mac · Coding help

Best pick
Tight Qwen3.8-27B qwen3.8:27b · 18 GB download · AA index 28 (medium, Ollama's default thinking level)
Highest independent score among the models we track that run on this machine. It is tight here: short chats are fine, long chats may slow down. For quick answers, type /set nothink in the chat.
```
ollama run qwen3.8:27b
```
Alternative
Fits Gemma 4 (26B-A4B / 31B) gemma4:e4b · 9.5 GB download · AA index 9 (E4B)
Why consider it: Fits with room for long coding sessions. Qwen3.8 is tight on a 32 GB Mac, but scores much higher. Every Gemma 4 size reads images.
```
ollama run gemma4:e4b
```

Also fits, but has no independent score yet so we can't rank it: LFM2.5-24B-A2B (lfm2.5:24b).

32 GB Mac · Reading images

Best pick
Tight Qwen3.8-27B qwen3.8:27b · 18 GB download · AA index 28 (medium, Ollama's default thinking level)
Highest independent score among the models we track that run on this machine. It is tight here: short chats are fine, long chats may slow down. For quick answers, type /set nothink in the chat.
```
ollama run qwen3.8:27b
```
Alternative
Fits Gemma 4 (26B-A4B / 31B) gemma4:e4b · 9.5 GB download · AA index 9 (E4B)
Why consider it: Fits with room to spare. Qwen3.8 is tight on a 32 GB Mac, but scores much higher. Every Gemma 4 size reads images.
```
ollama run gemma4:e4b
```

Also fits, but has no independent score yet so we can't rank it: MiniCPM-V 4.7 (35B-A3B) (openbmb/minicpm-v4.7:1b).

32 GB Mac · Long documents

Best pick
Fits Gemma 4 (26B-A4B / 31B) gemma4:e4b · 9.5 GB download · AA index 9 (E4B)
Highest independent score among the models we track that fit this machine. Every Gemma 4 size reads images.
```
ollama run gemma4:e4b
```

64 GB+ Mac · Everyday chat

Best pick
Fits Qwen3.8-27B qwen3.8:27b · 18 GB download · AA index 28 (medium, Ollama's default thinking level)
Highest independent score among the models we track that fit this machine. For quick answers, type /set nothink in the chat.
```
ollama run qwen3.8:27b
```

Also fits, but has no independent score yet so we can't rank it: MiniCPM-V 4.7 (35B-A3B) (openbmb/minicpm-v4.7:1b), LFM2.5-24B-A2B (lfm2.5:24b).

64 GB+ Mac · Coding help

Best pick
Fits Qwen3.8-27B qwen3.8:27b · 18 GB download · AA index 28 (medium, Ollama's default thinking level)
Highest independent score among the models we track that fit this machine. For quick answers, type /set nothink in the chat.
```
ollama run qwen3.8:27b
```

Also fits, but has no independent score yet so we can't rank it: LFM2.5-24B-A2B (lfm2.5:24b).

64 GB+ Mac · Reading images

Best pick
Fits Qwen3.8-27B qwen3.8:27b · 18 GB download · AA index 28 (medium, Ollama's default thinking level)
Highest independent score among the models we track that fit this machine. For quick answers, type /set nothink in the chat.
```
ollama run qwen3.8:27b
```

Also fits, but has no independent score yet so we can't rank it: MiniCPM-V 4.7 (35B-A3B) (openbmb/minicpm-v4.7:1b).

64 GB+ Mac · Long documents

Best pick
Fits Qwen3.8-27B qwen3.8:27b · 18 GB download · AA index 28 (medium, Ollama's default thinking level)
Highest independent score among the models we track that fit this machine. For quick answers, type /set nothink in the chat.
```
ollama run qwen3.8:27b
```

AMD Ryzen AI Max+ PC, 128 GB (graphics memory set to 96 GB) · Everyday chat

Best pick
Fits Qwen3.8-27B qwen3.8:27b · 18 GB download · AA index 28 (medium, Ollama's default thinking level)
Highest independent score among the models we track that fit this machine. For quick answers, type /set nothink in the chat.
```
ollama run qwen3.8:27b
```

Also fits, but has no independent score yet so we can't rank it: MiniCPM-V 4.7 (35B-A3B) (openbmb/minicpm-v4.7:1b), LFM2.5-24B-A2B (lfm2.5:24b), Laguna XS 2.1 (33B-A3B) (laguna-xs-2.1).

AMD Ryzen AI Max+ PC, 128 GB (graphics memory set to 96 GB) · Coding help

Best pick
Fits Qwen3.8-27B qwen3.8:27b · 18 GB download · AA index 28 (medium, Ollama's default thinking level)
Highest independent score among the models we track that fit this machine. For quick answers, type /set nothink in the chat.
```
ollama run qwen3.8:27b
```

Also fits, but has no independent score yet so we can't rank it: LFM2.5-24B-A2B (lfm2.5:24b), Laguna XS 2.1 (33B-A3B) (laguna-xs-2.1).

AMD Ryzen AI Max+ PC, 128 GB (graphics memory set to 96 GB) · Reading images

Best pick
Fits Qwen3.8-27B qwen3.8:27b · 18 GB download · AA index 28 (medium, Ollama's default thinking level)
Highest independent score among the models we track that fit this machine. For quick answers, type /set nothink in the chat.
```
ollama run qwen3.8:27b
```

Also fits, but has no independent score yet so we can't rank it: MiniCPM-V 4.7 (35B-A3B) (openbmb/minicpm-v4.7:1b).

AMD Ryzen AI Max+ PC, 128 GB (graphics memory set to 96 GB) · Long documents

Best pick
Fits Qwen3.8-27B qwen3.8:27b · 18 GB download · AA index 28 (medium, Ollama's default thinking level)
Highest independent score among the models we track that fit this machine. For quick answers, type /set nothink in the chat.
```
ollama run qwen3.8:27b
```

Also fits, but has no independent score yet so we can't rank it: Laguna XS 2.1 (33B-A3B) (laguna-xs-2.1).

Picks rebuilt 2026-10-19 from scores checked 2026-10-19. Speeds are not compared, and we never publish benchmark numbers we did not source.