← Back to directory
OpenAI
o1
Released Dec 5, 2024 · knowledge cutoff 2023-10
- Status
- Superseded
- Location
- United States
- Modality
- Multimodal
- Context window
- 200K
- Max output
- 100K
- Speed
- —
- Price ($/MTok in / out)
- $15 / $60
- Cost per task
- —
- Open weights
- No
Benchmarks5 of 8 reported
GPQA Diamond
78%
SWE-bench Verified
48.9%
AIME (latest)
79.2%
Humanity's Last Exam
9.1%
LMArena Elo
1352
MMLU-Pro
—
Terminal-Bench 2.1
—
ARC-AGI-2
—
Strengths & weaknesses
Strengths
- First model that would genuinely stop and work a problem through before answering
- Catches and reverses its own wrong turns mid-reasoning instead of committing to the first path
- Best on problems with a checkable answer — proofs, puzzles, tight logic
Weaknesses
- Long, unpredictable wait before the first token; unusable anywhere interactive
- Overthinks trivial questions, spending minutes on what a small model answers instantly
- Reasoning is hidden, so you cannot see where it went wrong or steer it
Full release of the o1 series (previewed September 2024).
For developers
API model strings
Not researched
Licence
Proprietary — weights not released
Retirement
No retirement announced
Lineage
No recorded predecessor