Metr time horizons — Epoch AI

Metr time horizons — Epoch AI is run by Epoch AI. It has ranked 50 models, of which the catalogue holds 35, scoring from 0.101 to 0.789.

measured by Epoch AI · the board itself

Place	Model	Metric	Score
2ndof 50	Claude Opus 4.6	average_score	0.789
3rdof 50	Gemini 3.1 Pro Preview	average_score	0.77
4thof 50	GPT-5.2	average_score (high)	0.753
5thof 50	Claude Opus 4.5	average_score	0.75
6thof 50	GPT-5.3 Codex	average_score	0.745
7thof 50	GPT-5.4	average_score	0.743
10thof 50	gemini-3-pro-preview	average_score	0.71
11thof 50	GPT-5.1 Codex Max	average_score (max)	0.708
12thof 50	GPT-5	average_score (medium)	0.696
14thof 50	Claude Sonnet 4.5	average_score	0.674
15thof 50	Claude Opus 4.1	average_score	0.668
17thof 50	o3	average_score (medium)	0.654
18thof 50	Claude Opus 4	average_score	0.639
19thof 50	o4-mini	average_score (medium)	0.639
21stof 50	Claude Sonnet 4	average_score	0.62
24thof 50	Claude 3.7 Sonnet	average_score	0.6
25thof 50	Kimi K2 Thinking	average_score	0.592
26thof 50	GPT OSS 120B	average_score	0.566
28thof 50	Gemini 2.5 Pro Preview 06-05	average_score	0.554
29thof 50	DeepSeek-R1 (0528)	average_score	0.538
30thof 50	DeepSeek-R1	average_score	0.519
31stof 50	o1	average_score (medium)	0.511
32ndof 50	DeepSeek-V3 0324	average_score	0.496
33rdof 50	DeepSeek-V3	average_score	0.474
34thof 50	Claude 3.5 Sonnet	average_score	0.452
36thof 50	GPT-4o (2024-11-20)	average_score	0.408
38thof 50	GPT-4 Turbo	average_score	0.367
40thof 50	Qwen2.5 72B Instruct	average_score	0.358
42ndof 50	GPT-4o (2024-08-06)	average_score	0.338
43rdof 50	Qwen2-72B-Instruct	average_score	0.299
44thof 50	Claude 3 Opus	average_score	0.295
45thof 50	GPT-4 (0613)	average_score	0.293
48thof 50	GPT-3.5 Turbo Instruct	average_score	0.215
49thof 50	davinci-002	average_score	0.162
50thof 50	gpt2-xl	average_score	0.101

Place

Model

Metric

Score

2ndof 50

average_score

0.789

3rdof 50

average_score

0.77