← Back to directory
OpenAI
GPT-4
Released Mar 14, 2023 · knowledge cutoff 2021-09
- Status
- Deprecated
- Location
- United States
- Modality
- Multimodal
- Context window
- 33K
- Max output
- 8K
- Speed
- —
- Price ($/MTok in / out)
- $30 / $60
- Cost per task
- —
- Open weights
- No
Benchmarks2 of 8 reported
GPQA Diamond
35.7%
LMArena Elo
1163
MMLU-Pro
—
SWE-bench Verified
—
Terminal-Bench 2.1
—
AIME (latest)
—
Humanity's Last Exam
—
ARC-AGI-2
—
Strengths & weaknesses
Strengths
- Literal, dependable instruction follower — did what you asked and no more
- Held a consistent persona and thread across a long conversation, which earlier models lost within a few turns
- Broad enough general knowledge to be a default assistant for almost any subject
Weaknesses
- Answers in one pass — no deliberation, so hard problems fail instantly and confidently
- Fabricates citations, APIs and quotes with no signal that it is guessing
- Slow token-by-token output that felt sluggish even at the time
The model that defined the 2023 frontier.
For developers
API model strings
Not researched
Licence
Proprietary — weights not released
Retirement
Retirement date not researched
Lineage
No recorded predecessor