Skip to content

Benchmark

METR Time Horizons

Agentic benchmarks measure how well models carry out tasks that take several steps.

Models measured
35
Scores recorded
41
Sources
1
IQ weight
2.99

Score distribution

120.5
Mean
74.9
Median
146.6
Std dev
4
10th percentile
341.7
90th percentile
0.144
Min
0.144718.8
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.99 for chat models, measured over 35 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Claude Opus 4.6· Epoch AI Benchmarking Hub data718.8
2Gemini 3.1 Pro Preview· Epoch AI Benchmarking Hub data384.1
3GPT 5.2high· Epoch AI Benchmarking Hub data352.2
4GPT-5.3 Codex· Epoch AI Benchmarking Hub data349.5
5GPT 5.4xhigh· Epoch AI Benchmarking Hub data341.7
6Claude Opus 4.5· Epoch AI Benchmarking Hub data293
7Claude Opus 4.516k· Epoch AI Benchmarking Hub data288.9
8Gemini 3 Pro Preview· Epoch AI Benchmarking Hub data224.3
9GPT-5.1 Codex Max· Epoch AI Benchmarking Hub data223.7
10GPT 5high· Epoch AI Benchmarking Hub data203
11GPT 5medium· Epoch AI Benchmarking Hub data137.3
12Claude Sonnet 4.516k· Epoch AI Benchmarking Hub data121.9
13OpenAI o3· Epoch AI Benchmarking Hub data119.7
14Claude Opus 4.116k· Epoch AI Benchmarking Hub data113.7
15Grok 4· Epoch AI Benchmarking Hub data110.1
16Claude Opus 4.1· Epoch AI Benchmarking Hub data100.5
17Claude Opus 4· Epoch AI Benchmarking Hub data100.4
18OpenAI o3medium· Epoch AI Benchmarking Hub data91.3
19Claude Opus 416k· Epoch AI Benchmarking Hub data85.6
20o4 Minimedium· Epoch AI Benchmarking Hub data76.5
21Claude Sonnet 416k· Epoch AI Benchmarking Hub data74.9
22Claude Sonnet 3.7· Epoch AI Benchmarking Hub data60.4
23Claude Sonnet 3.716k· Epoch AI Benchmarking Hub data56.1
24Kimi K2 Thinking· Epoch AI Benchmarking Hub data54.2
25GPT OSS 120B· Epoch AI Benchmarking Hub data42
26o1medium· Epoch AI Benchmarking Hub data38.8
27Gemini 2.5 Pro· Epoch AI Benchmarking Hub data38.7
28DeepSeek R1· Epoch AI Benchmarking Hub data31.2
29DeepSeek V3· Epoch AI Benchmarking Hub data23.1
30Claude 3.5 Sonnet· Epoch AI Benchmarking Hub data20.5
31O1 Preview· Epoch AI Benchmarking Hub data20.3
32GPT 4o· Epoch AI Benchmarking Hub data9.2
33GPT-4 Turbo· Epoch AI Benchmarking Hub data6.6
34GPT-4-0125-preview· Epoch AI Benchmarking Hub data5.4
35GPT 4· Epoch AI Benchmarking Hub data5.4
36Qwen2.5 72B Instruct· Epoch AI Benchmarking Hub data5.2
37GPT 4 1106 Preview· Epoch AI Benchmarking Hub data4
38Claude 3 Opus· Epoch AI Benchmarking Hub data4
39Qwen2 72B Instruct· Epoch AI Benchmarking Hub data2.2
40GPT 35 Turbo Instruct· Epoch AI Benchmarking Hub data0.599
41Davinci 002· Epoch AI Benchmarking Hub data0.144

Data credits: Epoch AI Benchmarking Hub, CC BY 4.0 (licence)