intelligence index

Every model, measured.

A leaderboard of open and proprietary models scored on real hardware. The models flagged local run entirely on-device through Aperture — the same set the router prefers first.

56 / 56 models
#model
1
Claude Fable 5
anthropic · · 500k · proprietary
65
65
64
75
43
2
Claude Opus 4.8
anthropic · · 500k · proprietary
61
67
54
65
39
3
GPT-5.5
openai · · 400k · proprietary
60
58
56
62
40
4
Gemini 3 Ultra
google · · 1000k · proprietary
59
61
47
65
45
5
GPT-5
openai · · 400k · proprietary
58
57
56
61
43
6
DeepSeek R2
deepseek · 671B · 128k · open
57
61
48
61
37
7
Claude Sonnet 5
anthropic · · 500k · proprietary
56
55
54
55
40
8
Gemini 3 Pro
google · · 1000k · proprietary
55
55
55
56
40
9
Grok 4
xai · · 256k · proprietary
54
57
43
63
38
10
Qwen3 Max
alibaba · 480B · 256k · open
53
59
47
56
37
11
DeepSeek V3.2
deepseek · 671B · 128k · open
52
52
53
51
40
12
Llama 4 Behemoth
meta · 405B · 256k · open
51
49
49
54
34
13
Kimi K2
moonshot · 1T · 256k · open
50
52
53
58
38
14
Qwen3 235B
alibaba · 235B · 128k · open
49
51
37
57
36
15
Mistral Large 3
mistral · 123B · 128k · open
48
47
47
56
36
16
GPT-5 mini
openai · · 400k · proprietary
47
49
51
51
33
17
Llama 4 Maverick
meta · 400B · 256k · open
47
48
50
50
34
18
Claude Haiku 4.5
anthropic · · 200k · proprietary
46
48
50
47
29
19
GLM-5
zhipu · 355B · 200k · open
46
42
41
52
31
20
Ernie 5
baidu · · 128k · proprietary
45
43
43
46
31
21
Gemini 3 Flash
google · · 1000k · proprietary
45
46
48
45
34
22
DeepSeek V3.2 Lite
deepseek · 70B · 128k · open
44
41
41
46
33
23
MiniMax M2
minimax · · 1000k · proprietary
44
48
37
42
30
24
Nova Pro 2
amazon · · 300k · proprietary
43
44
46
42
30
25
Qwen3 72Blocal
alibaba · 72B · 128k · open
43
48
37
52
29
26
Command R+ 2
cohere · 104B · 128k · open
42
43
45
47
32
27
Hunyuan Large 2
tencent · 389B · 256k · open
42
46
34
42
30
28
Jamba 2 Large
ai21 · 398B · 256k · open
41
41
43
39
26
29
Llama 4 Scoutlocal
meta · 109B · 512k · open
41
45
33
40
29
30
Grok 4 mini
xai · · 128k · proprietary
40
46
35
37
27
31
Qwen3 32Blocal
alibaba · 32B · 128k · open
39
40
42
39
30
32
DBRX 2
databricks · 132B · 64k · open
38
36
36
38
24
33
Gemma 3 27Blocal
google · 27B · 128k · open
37
39
42
39
26
34
Arctic 2
snowflake · 480B · 64k · open
36
38
26
44
22
35
Mistral Small 3.2local
mistral · 24B · 128k · open
35
39
28
38
27
36
Command R 2local
cohere · 35B · 128k · open
34
35
38
38
26
37
Nova Lite 2
amazon · · 300k · proprietary
34
35
38
41
26
38
OLMo 3 32Blocal
allenai · 32B · 65k · open
33
32
34
38
21
39
Qwen3 14Blocal
alibaba · 14B · 128k · open
33
36
25
33
24
40
Phi-5local
microsoft · 14B · 128k · open
32
33
37
36
19
41
Yi-2 34Blocal
01ai · 34B · 200k · open
32
30
31
40
19
42
Reka Flash 3local
reka · 21B · 128k · open
31
31
34
31
24
43
Aya Expanse 2local
cohere · 32B · 128k · open
30
32
36
38
17
44
Falcon 3 40Blocal
tii · 40B · 32k · open
30
26
27
34
23
45
Gemma 3 12Blocal
google · 12B · 128k · open
30
27
28
36
23
46
Qwen3 8Blocal
alibaba · 8B · 128k · open
29
28
31
33
23
47
Llama 3.3 8Blocal
meta · 8B · 128k · open
28
25
26
27
17
48
Ministral 8Blocal
mistral · 8B · 128k · open
26
30
20
27
21
49
Gemma 3 4Blocal
google · 4B · 128k · open
24
29
20
32
15
50
Qwen3 4Blocal
alibaba · 4B · 32k · open
23
26
16
26
18
51
Phi-5 minilocal
microsoft · 3.8B · 128k · open
22
21
25
29
15
52
Llama 3.2 3Blocal
meta · 3B · 128k · open
21
20
23
25
18
53
SmolLM3 3Blocal
huggingface · 3B · 64k · open
19
19
23
15
14
54
Qwen3 1.7Blocal
alibaba · 1.7B · 32k · open
17
13
15
23
12
55
Gemma 3 1Blocal
google · 1B · 32k · open
16
16
21
14
13
56
Llama 3.2 1Blocal
meta · 1B · 128k · open
15
10
13
14
14

real-hardware benchmarks · component scores derived for demo · sort by any column