← Back to directory
OpenAI
o3
Released Apr 16, 2025 · knowledge cutoff 2024-06
- Status
- Deprecated
- Location
- United States
- Modality
- Multimodal
- Context window
- 200K
- Max output
- 100K
- Speed
- —
- Price ($/MTok in / out)
- $2 / $8
- Cost per task
- —
- Open weights
- No
Benchmarks7 of 8 reported
MMLU-Pro
85.6%
GPQA Diamond
83.3%
SWE-bench Verified
69.1%
AIME (latest)
91.6%
Humanity's Last Exam
20.3%
LMArena Elo
1418
ARC-AGI-2
6.5%
Terminal-Bench 2.1
—
Strengths & weaknesses
Strengths
- Thinks with tools in the loop — searches, runs code, then reasons over what came back
- Stays with a hard problem across many steps instead of settling early
- Reasoning effort is tunable, so one model covers quick and deep work
Weaknesses
- Invents confident specifics about its own actions, including claiming it ran code it never ran
- Latency swings wildly with effort — the same prompt takes seconds or minutes
- Clipped, telegraphic prose that reads as brusque after the 4o line
Pricing reflects the June 2025 cut from $10/$40. OpenAI is retiring o3 from ChatGPT on 2026-08-26 following a 90-day sunset notice; marked deprecated ahead of that date. API availability after the ChatGPT retirement was not specified in the announcement.
For developers
API model strings
Not researched
Licence
Proprietary — weights not released
Retirement
Retirement date not researched
Lineage
No recorded predecessor