Build your own index

Every intelligence index is a weighted average of benchmark scores, and somebody picked the weights. Here, you pick them. See who comes out on top, and whether the ranking holds when the weights shift.

Start from

The benchmarks the most labs reported in 2026 (32 labs down to 20), minus any where top models already score past 92%. We use the version most recent models were scored on.

Tests and weights

Pick from 1,158 benchmarks with at least 8 scored models.

Terminal-Bench 2.162 models13%
Humanity's Last Exam180 models13%
SWE-Bench Pro97 models13%
LiveCodeBench v6108 models13%
DeepSWE67 models13%
MMLU-Pro293 models13%
AutomationBench70 models13%
MCP Atlas63 models13%

Scores from

Independent runs Labs' own numbers Both

We use independent runs where they exist and fill the gaps with labs' own numbers.

A model needs scores on at least of the tests every test

Where a model skipped a test, we predict its score from its other results and how hard that test was.

Who comes out on top?

Claude Fable 5.2 leads this mix with 82.7 and stays first under every nearby weighting. Claude Opus 5.5 is second at 79.2. We rank the 73 models with scores on at least 4 of the 8 tests, from independent runs and labs’ own numbers.

82.7Claude Fable 5.2
79.2Claude Opus 5.5
77.3GLM-5.3
77.1LongCat-2.5
76.8DeepSeek V4.1 Flash
76.8GPT-6 Astra
76.0Inkling 2 Small
75.8Step 5 Preview
75.7GLM-5.3-Flash
75.5Claude Sonnet 5.5
75.1Claude Opus 5
74.3DeepSeek V4 Pro 0813
73.6Gemini 3.8 Flash
73.0GPT-5.6 Sol
73.0Claude Fable 5
72.3Kimi K3
71.7Muse Spark Open
71.6Claude Fable 5.1
71.4Tencent Hy4 preview
70.9Ling-3.1-flash
70.7Qwen 3.8 Max
69.5GPT-5.6 Terra
68.6Muse Spark 1.1
68.5Claude Opus 4.8
67.6Beam 1.0
67.0GPT-5.6 Luna
66.5GPT-5.5
65.1Claude Sonnet 5
64.3Qwen3.8-Omni-Flash
64.0Qwen3.8 Flash
63.8GLM-5.2
63.8DeepSeek-V4-Pro
62.9Gemini 3.6 Flash
62.7Qwen3.7-Max
62.4Seed2.1 Pro
61.2Inkling-Small
60.9Claude Opus 4.7
60.7Tencent Hy3
60.7Gemini 3.5 Flash
59.8Qwen 3.8 27B

1Claude Fable 5.282.7
2Claude Opus 5.579.2
3GLM-5.377.3
4LongCat-2.577.1
5DeepSeek V4.1 Flash76.8
6GPT-6 Astra76.8
7Inkling 2 Small76.0
8Step 5 Preview75.8
9GLM-5.3-Flash75.7
10Claude Sonnet 5.575.5
11Claude Opus 575.1
12DeepSeek V4 Pro 081374.3
13Gemini 3.8 Flash73.6
14GPT-5.6 Sol73.0
15Claude Fable 573.0
16Kimi K372.3
17Muse Spark Open71.7
18Claude Fable 5.171.6
19Tencent Hy4 preview71.4
20Ling-3.1-flash70.9
21Qwen 3.8 Max70.7
22GPT-5.6 Terra69.5
23Muse Spark 1.168.6
24Claude Opus 4.868.5
25Beam 1.067.6
26GPT-5.6 Luna67.0
27GPT-5.566.5
28Claude Sonnet 565.1
29Qwen3.8-Omni-Flash64.3
30Qwen3.8 Flash64.0
31GLM-5.263.8
32DeepSeek-V4-Pro63.8
33Gemini 3.6 Flash62.9
34Qwen3.7-Max62.7
35Seed2.1 Pro62.4
36Inkling-Small61.2
37Claude Opus 4.760.9
38Tencent Hy360.7
39Gemini 3.5 Flash60.7
40Qwen 3.8 27B59.8
41Inkling59.7
42DeepSeek-V4-Flash59.2
43GPT-5.458.8
44GLM-5.158.6
45Seed2.0 Pro58.0
46Kimi K2.658.0
47Qwen 3.6 Plus57.6
48Claude Opus 4.656.7
49Nemotron 3.5 Super VL56.5
50Gemini 3 Pro56.4
51MiniMax-M355.7
52Gemini 3.1 Pro54.1
53Claude Sonnet 4.654.0
54Qwen3.7-Plus54.0
55Solar Open 252.5
56GLM-4.752.2
57Muse Glimmer52.1
58Gemini 3 Flash51.4
59Nemotron 3 Ultra51.3
60Kimi K2.551.3
61Claude Opus 4.551.1
62Qwen3.6 27B50.9
63Qwen 3.6 35B-A3B48.7
64GPT-5.247.8
65Claude Sonnet 4.545.4
66DeepSeek-V3.245.2
67MiniMax-M2.743.5
68Gemma 4 31B IT42.1
69LFM3-1.7B38.6
70Granite 4.2 30B37.1
71Claude Sonnet 436.1
72gpt-oss-120b31.0
73Kimi K230.5

Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards

Rank	Model	Index	Tests	Best	Worst	First
1	Claude Fable 5.2	82.7	4 of 8	1	1	100%
2	Claude Opus 5.5	79.2	5 of 8	2	3
3	GLM-5.3	77.3	5 of 8	3	5
4	LongCat-2.5	77.1	5 of 8	3	8
5	DeepSeek V4.1 Flash	76.8	5 of 8	2	19
6	GPT-6 Astra	76.8	4 of 8	3	8
7	Inkling 2 Small	76.0	4 of 8	5	11
8	Step 5 Preview	75.8	5 of 8	5	11
9	GLM-5.3-Flash	75.7	4 of 8	5	14
10	Claude Sonnet 5.5	75.5	4 of 8	5	14
11	Claude Opus 5	75.1	5 of 8	4	13
12	DeepSeek V4 Pro 0813	74.3	4 of 8	8	14
13	Gemini 3.8 Flash	73.6	4 of 8	9	18
14	GPT-5.6 Sol	73.0	5 of 8	7	18
15	Claude Fable 5	73.0	6 of 8	9	20
16	Kimi K3	72.3	5 of 8	14	20
17	Muse Spark Open	71.7	4 of 8	13	21
18	Claude Fable 5.1	71.6	5 of 8	14	21
19	Tencent Hy4 preview	71.4	6 of 8	16	21
20	Ling-3.1-flash	70.9	6 of 8	14	25
21	Qwen 3.8 Max	70.7	5 of 8	18	22
22	GPT-5.6 Terra	69.5	4 of 8	19	25
23	Muse Spark 1.1	68.6	5 of 8	21	27
24	Claude Opus 4.8	68.5	6 of 8	21	25
25	Beam 1.0	67.6	6 of 8	22	28
26	GPT-5.6 Luna	67.0	4 of 8	23	28
27	GPT-5.5	66.5	5 of 8	24	29
28	Claude Sonnet 5	65.1	5 of 8	26	31
29	Qwen3.8-Omni-Flash	64.3	4 of 8	28	35
30	Qwen3.8 Flash	64.0	4 of 8	29	36
31	GLM-5.2	63.8	6 of 8	28	34
32	DeepSeek-V4-Pro	63.8	6 of 8	28	39
33	Gemini 3.6 Flash	62.9	4 of 8	31	37
34	Qwen3.7-Max	62.7	4 of 8	29	38
35	Seed2.1 Pro	62.4	5 of 8	28	40
36	Inkling-Small	61.2	4 of 8	34	41
37	Claude Opus 4.7	60.9	5 of 8	33	43
38	Tencent Hy3	60.7	5 of 8	34	43
39	Gemini 3.5 Flash	60.7	6 of 8	34	45
40	Qwen 3.8 27B	59.8	5 of 8	36	47
41	Inkling	59.7	4 of 8	37	44
42	DeepSeek-V4-Flash	59.2	7 of 8	35	47
43	GPT-5.4	58.8	6 of 8	39	48
44	GLM-5.1	58.6	5 of 8	38	48
45	Seed2.0 Pro	58.0	4 of 8	36	53
46	Kimi K2.6	58.0	4 of 8	36	56
47	Qwen 3.6 Plus	57.6	6 of 8	42	50
48	Claude Opus 4.6	56.7	4 of 8	44	53
49	Nemotron 3.5 Super VL	56.5	6 of 8	42	54
50	Gemini 3 Pro	56.4	5 of 8	43	54
51	MiniMax-M3	55.7	4 of 8	45	57
52	Gemini 3.1 Pro	54.1	7 of 8	49	61
53	Claude Sonnet 4.6	54.0	5 of 8	48	60
54	Qwen3.7-Plus	54.0	5 of 8	50	61
55	Solar Open 2	52.5	4 of 8	49	62
56	GLM-4.7	52.2	4 of 8	51	62
57	Muse Glimmer	52.1	4 of 8	52	63
58	Gemini 3 Flash	51.4	4 of 8	52	63
59	Nemotron 3 Ultra	51.3	5 of 8	53	62
60	Kimi K2.5	51.3	5 of 8	55	61
61	Claude Opus 4.5	51.1	4 of 8	54	62
62	Qwen3.6 27B	50.9	4 of 8	55	64
63	Qwen 3.6 35B-A3B	48.7	5 of 8	61	65
64	GPT-5.2	47.8	4 of 8	62	66
65	Claude Sonnet 4.5	45.4	4 of 8	64	68
66	DeepSeek-V3.2	45.2	4 of 8	61	69
67	MiniMax-M2.7	43.5	4 of 8	63	70
68	Gemma 4 31B IT	42.1	4 of 8	66	70
69	LFM3-1.7B	38.6	5 of 8	64	72
70	Granite 4.2 30B	37.1	4 of 8	69	71
71	Claude Sonnet 4	36.1	4 of 8	69	72
72	gpt-oss-120b	31.0	4 of 8	70	73
73	Kimi K2	30.5	4 of 8	72	73

How much do the weights matter?

In the default mix, 7 of the top 10 models can move three places or more when the weights shift.

How it works

We put every test on a 0 to 100 scale. Percentages stay as they are. Elo and other units run from the lowest model to the highest, and tests where lower is better get flipped. A model's index is the weighted average.

Where a model skipped a test, we predict that score from its other results and the test's difficulty, so skipping a hard test doesn't help and taking one doesn't hurt. Hover a model to see which of its scores are predicted.

Why weights matter

Two indexes built from the same scores can disagree on who's first, because the weights decide which tests count most. Published indexes change their weights between versions, and every change reorders some models.

So we rerank every model under 400 weightings near yours, each weight moved up or down by as much as half. Rank range shows each model's best and worst place, and First is the share of those weightings it wins.

What it leaves out

Independent runs cover far fewer new models than labs' own posts, so the board starts with both. Switch to Independent runs to leave lab numbers out, though the newest models may drop off.

Predictions assume a model's strengths carry across tests, so they miss a model that's strong at one thing and weak at another. Raise the minimum to lean on them less. And no average says how sure to be about any one score.