OpenAI

o3

Multimodal
#2MMMU

by OpenAI

+
+
+
+
About

OpenAI o3, released in April 2025, applies extended reinforcement learning training to produce a reasoning model that tackles complex problems across math, science, and coding. Its strong results on frontier benchmarks generated significant industry discussion about the scaling trajectory of RL-based reasoning models, and it represented OpenAI's investment in compute-heavy reasoning training as the primary lever for advancing capability.

+
+
+
+
Pricing Range
Input (per 1M)$10.00 -$10.00
Output (per 1M)$40.00 -$40.00
Providers1
+
+
+
+
Timeline
ReleasedApr 16, 2025
+
+
+
+
Specifications
Capabilities
Multimodal
+
+
+
+
License & Family
License
Proprietary
Performance Overview
Performance metrics and category breakdown

Overall Performance

5 benchmarks
Average Score
38.6%
Best Score
76.4%
High Performers (80%+)
0

Performance Metrics

Max Context Window
300.0K

Top Categories

Multimodal
76.4%
Tool Use
43.6%
Agents
26.9%
Reasoning
19.2%
+
+
+
+
All Benchmark Results for o3
Complete list of benchmark scores with detailed information
MMMU
Multimodal
76.40
76.4%
Unverified
MCP-Atlas
Tool Use
43.60
43.6%
Unverified
GDPVal
Agents
30.80
30.8%
Unverified
OSWorld
Agents
23.00
23.0%
Unverified
Humanity's Last Exam
Reasoning
19.20
19.2%
Unverified
+
+
+
+
Resources