Scores refreshed Oct 9Epoch AI

Every model, ranked.
Scores, prices and context.

337 models from 46 labs. 274 are independently scored; the rest show up as soon as they are.

Leaders

Top model

ECI 167.3

Claude Opus 5.5

Best open weights

#15 overall

Kimi K3

Cheapest in the top 25

$0.10 per 1M input tokens

GPT-6 Luna

Newest listed

StepFun

Step 5 Preview

Each lab's best

Over time

The frontier keeps moving.

Open weights 104
Closed 119
Best so far

Figure 1. Every scored model by release date and Capabilities Index. The line is the best score so far.

Every model

All 337.

AI models with their Capabilities Index, benchmark scores and input price
#
1	Claude Opus 5.5Anthropic	167.3	87.5	91.2	93.3	$4.00
2	GPT-6 AstraOpenAI	166.4	94.4	93.7	95.0	$10
3	GPT-6.1 SolOpenAI	166.1	93.9	93.7	94.2	$2.00
4	Claude Sonnet 5.5Anthropic	165.0	94.1	88.8	–	$2.00
5	Claude Fable 5.1Anthropic	164.7	–	90.2	90.0	$10
6	Claude Opus 5Anthropic	162.8	91.8	85.6	90.4	$5.00
7	GPT-6 SolOpenAI	162.7	92.4	89.8	89.6	$2.00
8	GPT-5.5 ProOpenAI	162.1	91.9	87.7	84.6	$30
9	Claude Fable 5Anthropic	162.1	81.1	87.0	89.2	$10
10	GPT-5.6 SolOpenAI	161.7	91.3	89.1	92.5	$2.00
11	GPT-5.6 TerraOpenAI	159.6	91.1	86.0	83.9	$2.00
12	GPT-5.5OpenAI	159.1	92.0	85.3	85.0	$5.00
13	GPT-5.4 ProOpenAI	158.9	92.8	82.5	83.3	$30
14	Claude Opus 4.8Anthropic	158.2	88.0	80.0	72.1	$5.00
15	Kimi K3Moonshot · open	157.4	90.8	72.2	60.4	$0.80
16	Gemini 3.7 FlashGoogle	157.3	93.1	71.6	84.6	$0.75
17	GPT-5.4OpenAI	156.8	91.1	78.6	74.0	$2.50
18	GPT-5.3 CodexOpenAI	156.8	–	–	–	$1.75
19	Muse Spark 1.3Meta	156.8	–	74.4	–	$1.25
20	Gemini 3.8 FlashGoogle	156.7	93.9	68.4	–	$0.75
21	Grok 4.6xAI	156.4	92.0	66.0	67.1	$2.00
22	Qwen 3.8 MaxAlibaba	156.4	90.2	74.7	–	–
23	GPT-5.6 LunaOpenAI	156.4	88.8	82.1	59.5	$0.20
24	GPT-6 LunaOpenAI	156.3	87.3	79.0	59.3	$0.10
25	Claude Opus 4.7Anthropic	156.3	86.9	70.2	75.8	$5.00
26	Claude Sonnet 5Anthropic	156.2	87.4	65.6	–	$2.00
27	GLM-5.3Z.ai	155.6	87.9	68.8	–	$0.039
28	GPT-5.2 ProOpenAI	155.4	–	74.0	54.2	$21
29	DeepSeek V4 Pro 0813DeepSeek · open	155.3	88.9	64.6	61.3	$0.66
30	Claude Opus 4.6Anthropic	155.2	87.4	66.0	69.2	$5.00
31	Qwen3.8 Max (0902)Alibaba	155.1	89.7	65.6	–	$2.00
32	DeepSeek V4.1 FlashDeepSeek · open	154.9	–	–	–	$0.30
33	Muse Spark 1.2Meta	154.9	–	–	–	$1.25
34	Gemini 3.1 ProGoogle	154.8	92.6	59.7	77.1	–
35	DeepSeek V4 Flash 0731DeepSeek · open	154.5	88.0	57.5	61.4	$0.006
36	Gemini 3.5 FlashGoogle	154.5	90.4	62.8	72.1	$1.50
37	Gemini 3.6 FlashGoogle	154.3	92.2	59.0	60.4	$0.75
38	Muse Spark 1.1Meta	154.2	–	–	–	$1.25
39	Grok 4.5xAI	153.9	91.3	57.2	52.6	$2.00
40	Qwen3.7-MaxAlibaba	153.7	87.9	64.6	–	$1.48
41	Grok 4.7xAI	153.5	90.3	53.0	–	$2.00
42	GPT-5.2OpenAI	153.4	88.5	67.4	52.9	$1.75
43	Gemini 3 ProGoogle	152.9	90.1	–	31.1	–
44	Claude Sonnet 4.6Anthropic	152.2	83.2	–	60.4	$3.00
45	Muse SparkMeta	152.0	86.4	–	–	–
46	Grok 4.20xAI	152.0	85.8	44.9	65.1	$1.25
47	GLM-5.3-FlashZ.ai · open	151.9	86.9	55.8	–	$0.15
48	Gemini 3 FlashGoogle	151.8	85.9	51.2	33.6	–
49	GLM-5.2Z.ai · open	151.8	89.1	59.2	22.8	$0.06
50	Kimi K2.6Moonshot · open	151.1	87.7	57.2	–	$0.47

337 models · scores in % · not yet scored at the bottom

Where the numbers come from

Scores are Epoch AI's own runs and the public leaderboards they collect, not launch claims. The Capabilities Index puts them on one scale. Prices and context windows are OpenRouter's listing.

GPQA Diamond: Graduate-level science questions written so that searching the web doesn't help.
SWE-bench Verified: Fixing real GitHub issues in Python projects; the human-checked subset.
FrontierMath: Unpublished research-level maths problems (tiers 1–3).
ARC-AGI-2: Abstract visual puzzles that are easy for people and hard for models.
Humanity's Last Exam: Expert-written questions across dozens of fields.
SimpleQA Verified: Short factual questions: does the model know, or make it up.
Mock AIME: Competition maths in the style of the AIME.
Terminal-Bench: Getting real tasks done in a command-line terminal.