o3

o3 was released April 16, 2025 as OpenAI's most powerful reasoning model. Scored 99.2% on MATH, 87.5% on ARC-AGI, and 83.3% on GPQA Diamond. Featured improved tool use, image understanding, and code execution. Priced at \$10/\$40 per million tokens with a 200K context window.

Released

2025-04-16

Context Window

200K tokens

Pricing

Type	Price per MTok
Input	$10.00
Output	$40.00

Benchmarks12

General

1 benchmark

93.2%#1/10

95th percentile

Knowledge

2 benchmarks

92%#4/37

91th percentile

47.6%#2/3

50th percentile

Reasoning

3 benchmarks

Humanity's Last Exam

19.8%#1/7

93th percentile

83.3%#4/34

90th percentile

87.5%#1/2

75th percentile

Math

3 benchmarks

99.2%#1/31

98th percentile

96.7%#1/12

96th percentile

85.3%#5/7

36th percentile

Coding

3 benchmarks

71.7%#2/9

83th percentile

Codeforces Rating

2,706#2/4

63th percentile

SWE-bench Verified

69.1%#8/18

58th percentile

Percentile among tested models:Top 25%50-75%25-50%Bottom 25%

o-series Family5

Model	Tier	Released	Input $/MTok
o4-mini	mini	2025-04-16	$1.10
o3-mini	mini	2025-01-31	$1.10
o1	flagship	2024-12-05	$15.00
o1-preview	preview	2024-09-12	$15.00
o1-mini	mini	2024-09-12	$3.00

Other OpenAI Models8

GPT-4.12025-04-14

GPT (flagship)

GPT-4.1 mini2025-04-14

GPT (mini)

GPT-4.1 nano2025-04-14

GPT (nano)

GPT-4o mini2024-07-18

GPT (mini)

GPT-4o2024-05-13

GPT (flagship)

GPT-4 Turbo2023-11-06

GPT (flagship)

GPT-42023-03-14

GPT (flagship)

GPT-3.5 Turbo2023-03-01

GPT (mid)

Details

Model Familyo-series

Tierflagship

Generationo3

Release Date2025-04-16

Context Window200K tokens

Open WeightNo

Modalitytext, image

Capabilities4

reasoningtool-usevisioncode-execution

Sources1

Introducing OpenAI o3 and o4-mini

Tags

openaireasoningo-seriesfrontier-ai