← Back to directory
OpenAI
GPT-4o
Released May 13, 2024 · knowledge cutoff 2023-10
- Status
- Superseded
- Location
- United States
- Modality
- Multimodal
- Context window
- 128K
- Max output
- 16K
- Speed
- —
- Price ($/MTok in / out)
- $2.50 / $10
- Cost per task
- —
- Open weights
- No
Benchmarks6 of 8 reported
MMLU-Pro
74.7%
GPQA Diamond
53.6%
SWE-bench Verified
33.2%
AIME (latest)
13.4%
Humanity's Last Exam
2.7%
LMArena Elo
1318
Terminal-Bench 2.1
—
ARC-AGI-2
—
Strengths & weaknesses
Strengths
- Conversational latency low enough for real-time voice — it can be interrupted and talk back
- Reads images, handwriting and screen captures directly rather than through a caption model
- Warm, chatty register that a lot of users still preferred to every model that replaced it
Weaknesses
- Agreeable to a fault — follows a user's wrong premise rather than correcting it
- Answers immediately, so multi-step problems fail in ways it has no chance to catch
- Shallow on anything needing a sustained derivation
OpenAI's first natively multimodal flagship.
For developers
API model strings
Not researched
Licence
Proprietary — weights not released
Retirement
No retirement announced
Lineage
No recorded predecessor