Ability of AI models to execute highly complex tasks has grown rapidly
Maximum complexity of a given task that an AI model can complete successfully*. Complexity measured by number of hours a human would take to complete the same task
Claude Mythos
preview
16 human hours
12
Claude Opus 4.6
Confidence
interval
8
GPT-3.5
High
4
GPT-2
launched in 2019
GPT-3.5
o3
GPT-4
0
2020
2022
2024
2026
AI model launch date
Claude Mythos
preview
16 human hours
12
Claude Opus 4.6
Confidence
interval
8
GPT-3.5
High
4
GPT-2
launched in 2019
GPT-3.5
o3
GPT-4
0
2020
2022
2024
2026
AI model launch date
Claude Mythos
preview
16 human hours
12
Claude Opus 4.6
Confidence
interval
8
GPT-3.5
High
4
GPT-2
launched in 2019
o3
GPT-3.5
GPT-4
0
AI model
launch date
2020
2022
2024
2026
Claude Mythos
preview
16 human hours
12
Claude Opus 4.6
Confidence
interval
8
GPT-3.5
High
4
GPT-2
launched in 2019
o3
GPT-3.5
GPT-4
0
AI model
launch date
2020
2022
2024
2026
Claude Mythos
preview
16 human hours
12
Claude Opus 4.6
Confidence
interval
8
GPT-3.5
High
4
GPT-2
launched in 2019
GPT-3.5
o3
GPT-4
0
2020
2022
2024
2026
AI model launch date
Claude Mythos
preview
16 human hours
12
Claude Opus 4.6
Confidence
interval
8
GPT-3.5
High
4
GPT-2
launched in 2019
GPT-3.5
o3
GPT-4
0
2020
2022
2024
2026
AI model launch date
Claude Mythos
preview
16 human hours
12
Claude Opus 4.6
Confidence
interval
8
GPT-3.5
High
4
GPT-2
launched in 2019
o3
GPT-3.5
GPT-4
0
AI model
launch date
2020
2022
2024
2026
Claude Mythos
preview
16 human hours
12
Claude Opus 4.6
Confidence
interval
8
GPT-3.5
High
4
GPT-2
launched in 2019
o3
GPT-3.5
GPT-4
0
AI model
launch date
2020
2022
2024
2026