Skip to content

The model index

All 114 open-weight models we track, grouped by size — because the first question is always whether it fits. Smallest VRAM footprint first within each tier.

Tinyunder 3B

10

Runs on almost anything — phones, a Raspberry Pi, any laptop.

Gemma 3 1B
1Btext generationmultilingualsummarization
Qwen 3.5 0.8B
800Mtext generationreasoningmultilingual
SmolLM2 1.7B
1.7Btext generationcode generationsummarization
DeepSeek R1 1.5B
1.5Btext generationreasoningmath
Llama 3.2 1B
1Btext generationsummarization
Qwen 3 0.6B
600Mtext generationreasoningmultilingual
Gemma 2 2B
2Btext generationcode generationmultilingual
Qwen 3.5 2B
2Btext generationcode generationreasoning
Gemma 3n E2B
2Btext generationreasoningmultilingual
Gemma 4 E2B
2Btext generationreasoningmultilingual

Small3–9B

30

Comfortable on 8 GB GPUs and entry-level Macs.

Llama 3.2 3B
3Btext generationcode generationmultilingual
StarCoder2 3B
3Bcode generationtext generation
Phi-3 Mini 3.8B
3.8Btext generationcode generationreasoning
Gemma 3n E4B
4Btext generationreasoningmultilingual
Phi-4 Mini 3.8B
3.8Btext generationcode generationreasoning
Qwen 3 4B
4Btext generationcode generationreasoning
Qwen 3.5 4B
4Btext generationcode generationreasoning
Gemma 3 4B
4Btext generationcode generationreasoning
Yi 1.5 6B
6Btext generationcode generationreasoning
InternLM 2.5 7B
7Btext generationcode generationreasoning
StarCoder2 7B
7Bcode generationtext generation
DeepSeek R1 7B
7Btext generationcode generationreasoning
Mistral 7B
7Btext generationcode generationmultilingual
Qwen 2.5 7B
7Btext generationcode generationmultilingual
Qwen 2.5 Coder 7B
7Btext generationcode generation
Dolphin 3 8B
8Btext generationcode generationreasoning
Gemma 4 E4B
4Btext generationcode generationreasoning
Granite 3.3 8B
8Btext generationcode generationreasoning
Nous Hermes 2 8B
8Btext generationcode generationreasoning
Llama 3.1 8B
8Btext generationcode generationmultilingual
Aya Expanse 8B
8Btext generationmultilingualcreative writing
Falcon 3 7B
7Btext generationcode generationreasoning
Codestral Mamba 7B
7Bcode generationtext generation
OpenChat 3.5 7B
7Btext generationcode generationreasoning
WizardLM 2 7B
7Btext generationcode generationreasoning
Qwen 2.5 VL 7B
7Btext generationvisionreasoning
Cogito 8B
8Btext generationcode generationreasoning
DeepSeek R1 8B
8Btext generationcode generationreasoning
Nemotron 3 Nano 8B
8Btext generationcode generationreasoning
Qwen 3 8B
8Btext generationcode generationreasoning

Medium9–32B

28

Wants a 16–24 GB card or a 32 GB+ Mac.

Yi 1.5 9B
9Btext generationcode generationreasoning
Gemma 2 9B
9Btext generationcode generationreasoning
Qwen 3.5 9B
9Btext generationcode generationreasoning
Yi Coder 9B
9Bcode generationtext generationreasoning
Falcon 3 10B
10Btext generationcode generationreasoning
Llama 3.2 Vision 11B
11Btext generationvisionreasoning
Mistral Nemo 12B
12Btext generationcode generationreasoning
DeepSeek R1 14B
14Btext generationcode generationreasoning
Phi-4 14B
14Btext generationcode generationreasoning
Qwen 2.5 14B
14Btext generationcode generationreasoning
Gemma 3 12B
12Btext generationcode generationreasoning
StarCoder2 15B
15Btext generationcode generation
Phi-4 Reasoning 14B
14Btext generationcode generationreasoning
InternLM 2.5 20B
20Btext generationcode generationreasoning
Qwen 2.5 Coder 14B
14Btext generationcode generation
Qwen 3 14B
14Btext generationcode generationreasoning
Codestral 22B
22Btext generationcode generation
gpt-oss 20B
21Btext generationcode generationreasoning
Devstral 24B
24Btext generationcode generationreasoning
Magistral Small 24B
24Btext generationcode generationreasoning
Gemma 2 27B
27Btext generationcode generationreasoning
Mistral Small 3.1 24B
24Btext generationcode generationreasoning
Qwen 3.5 27B
27Btext generationcode generationreasoning
Gemma 3 27B
27Btext generationcode generationreasoning
Gemma 4 26B
26Btext generationcode generationreasoning
Qwen 3.6 27B
27Btext generationcode generationreasoning
Gemma 4 31B
31Btext generationcode generationreasoning
Qwen 3 30B-A3B (MoE)
30Btext generationcode generationreasoning

Large32–70B

16

High-end GPUs (24–48 GB) or 64 GB+ unified memory.

Qwen 3.5 35B A3B
35Btext generationcode generationreasoning
Nous Hermes 2 34B
34Btext generationcode generationreasoning
DeepSeek R1 32B
32Btext generationcode generationreasoning
Qwen 2.5 32B
32Btext generationcode generationreasoning
Yi 1.5 34B
34Btext generationcode generationreasoning
Cogito 32B
32Btext generationcode generationreasoning
QwQ 32B
32Btext generationcode generationreasoning
Aya Expanse 32B
32Btext generationmultilingualcreative writing
Laguna XS 2.1
33Btext generationcode generationreasoning
WizardCoder 33B
33Bcode generationtext generationreasoning
Command R 35B
35Btext generationreasoningmultilingual
Qwen 2.5 Coder 32B
32Btext generationcode generationreasoning
Qwen 3 32B
32Btext generationcode generationreasoning
Dolphin Mixtral 8x7B
47Btext generationcode generationreasoning
Qwen 3.6 35B-A3B
35Btext generationcode generationreasoning
Mixtral 8x7B
47Btext generationcode generationreasoning

Huge70B and up

30

Workstation territory — multi-GPU or 96 GB+ Macs. MoE models cheat the rules.

Qwen 2.5 VL 72B
72Btext generationvisionreasoning
Cogito 70B
70Btext generationcode generationreasoning
DeepSeek R1 70B
70Btext generationcode generationreasoning
Llama 3.1 70B
70Btext generationcode generationreasoning
Llama 3.3 70B
70Btext generationcode generationreasoning
Qwen 2.5 72B
72Btext generationcode generationreasoning
Llama 3.2 Vision 90B
90Btext generationvisionreasoning
Command R+ 104B
104Btext generationcode generationreasoning
Command A 111B
111Btext generationcode generationreasoning
Devstral 2 123B
123Btext generationcode generationreasoning
Mistral Large 2 123B
123Btext generationcode generationreasoning
gpt-oss 120B
117Btext generationcode generationreasoning
Llama 4 Scout (109B/17B active)
109Btext generationcode generationreasoning
Qwen 3.5 122B
122Btext generationcode generationreasoning
Mixtral 8x22B
141Btext generationcode generationreasoning
Qwen 3 235B-A22B
235Btext generationcode generationreasoning
Nemotron Ultra 253B
253Btext generationcode generationreasoning
MiniMax M3
428Btext generationcode generationreasoning
Llama 4 Maverick
400Btext generationcode generationreasoning
Llama 3.1 405B
405Btext generationcode generationreasoning
GLM-5.2
753Btext generationcode generationreasoning
GLM-5
744Btext generationcode generationreasoning
GLM-5.1
754Btext generationcode generationreasoning
Kimi K2.7 Code
1Ttext generationcode generationreasoning
Kimi K2.6
1Ttext generationcode generationreasoning
DeepSeek R1 671B
671Btext generationcode generationreasoning
DeepSeek V3
671Btext generationcode generationreasoning
DeepSeek V3-0324
671Btext generationcode generationreasoning
Kimi K2.5
1Ttext generationcode generationreasoning
DeepSeek V3.2
671Btext generationcode generationreasoning