Compare Start here

OpenAILanguage modelSuperseded

OpenAI o3

The reasoning model that learned to think with images and tools.

Compare head-to-head Race it in 3D

Overall Intelligence

#42

of 56
19.1 pts behind #1

0

races led

323

weeks at #1

69.6

score

Released: Apr 16, 2025; 1.5 yr ago
Context: 200K; 100K max output
Price: $2 / $8; per 1M in/out · $3.50 blended
Speed: 60 tok/s; 12s to first token
Modalities: text · image; outputs text
Weights: Closed; API and apps

The 30-second version

The reasoning model that made tool use during thinking mainstream, and a 2025 leader on hard math and science.

Key takeaways

1Could search, run code and look at images while reasoning.
2Set records on hard reasoning benchmarks in spring 2025.

Where it ranks

Current snapshot · illustrative scores · click any race to watch it

o3Best on each axis

Overall

42Overall+19.1 pts

Coding

43Coding+21.8 pts 43Frontend & UI+130 Elo 43Backend & APIs+19.8 pts 42Distributed Systems+22.1 pts 46Full-Stack Apps+22.9 pts 44Debugging & Repair+22.8 pts 44Agentic Coding+24.2 pts 43Code Review & Security+19.4 pts 39Data Science & SQL+16.0 pts 45Mobile Apps+20.8 pts 43Terminal & DevOps+26.6 pts 38Competitive Programming+526

Hard Reasoning

37Hard Reasoning+18.2 pts 39Frontier Questions+31.5 pts 34Abstract & Novel Reasoning+24.1 pts 38Logic & Puzzles+15.7 pts 42Planning & Strategy+18.9 pts

Math

37Math+18.2 pts 35Competition Math+13.3 pts 38Research Math+23.8 pts 39Applied & Quantitative+16.0 pts

Science

40Science+18.4 pts 40Graduate Science QA+11.3 pts 41Biology & Medicine+16.8 pts 35Physics & Chemistry+13.9 pts 42Research Assistance+18.1 pts

Writing & Language

45Writing & Language+23.2 pts 48Creative Writing+152 Elo 48Technical Writing+19.2 pts 47Editing & Style+18.8 pts 45Multilingual+14.4 pts 42Instruction Following+14.8 pts 48Conversation & EQ+145 Elo

Multimodal

36Multimodal+29.6 pts 33Image Understanding+15.7 pts 36Documents & Charts+18.3 pts 36Video Understanding+30.7 pts 38Speech & Audio+39.3 pts

Tool Use & Agents

40Tool Use & Agents+20.9 pts 40Tool Use+15.7 pts 36Computer Use+27.9 pts 40Web Research+22.6 pts 45Long-Horizon+27.0 hrs

Long Context

43Long Context+24.9 pts 481M Recall+22.6 pts 40Long Reasoning+19.9 pts

Speed & Cost

52Speed & Cost+54.9 pts 45Output Speed+462 tok/s 55Latency+11.06s 44Cost-Efficiency+55.7 pts

Trust & Factuality

43Trust & Factuality+21.5 pts 43Factuality+17.8 pts 42Honesty & Calibration+17.6 pts

Overall Intelligence over time

#1 frontiero3V4.1 FlashGLM-5.3

Update history

The distilled version first; open any release for the deeper breakdown.

The OpenAI o-series family

o3-miniJanuary 2025 o3April 2025

60-second briefing

o3, explained

AI video summary · placeholder render from the auto-written script

From OpenAI's own guide, distilled

Prompting guides for GPT-5 and later models

Reasoning models do their own step-by-step thinking; give goals and constraints rather than scripting the chain of thought.
Use the reasoning-effort and verbosity controls instead of prompt tricks to trade depth for speed.
Contradictory instructions waste reasoning tokens; audit system prompts for conflicts.
For agents, define stop conditions and how eager the model should be to keep going without asking.
Structured outputs (JSON schema) beat asking politely for a format.

All official guides, distilled →

About OpenAI

Maker of ChatGPT and the GPT series, pursuing general-purpose AI at consumer scale.

Optimising for

Frontier reasoning and math, broad consumer usefulness inside ChatGPT, and tiered models (Astra, Sol, Luna) that trade depth for price and speed.

San Francisco, US · founded 2015

Closest rivals

o3 is 42nd in overall intelligence; leader is Claude Opus 5.5.