o3-mini

Name: o3-mini
Author: OpenAI

OpenAI · 2025-01-31

o3-mini is a small model in OpenAI's o-series of reasoning models, using chain-of-thought reasoning at lower cost and latency than the full o3. It is tuned for STEM tasks like math, science, and coding with selectable reasoning effort.

Benchmark scores

Benchmark	Score
AA-Briefcase Elo	161
AA-Briefcase v1.1 Elo	142
AA-LCR	39.3
AA-LCR v1.1	43.0
AA-Omniscience Index	-42.6
Agentic Tool Use (Chat)	63.5
Agentic Tool Use (Enterprise)	65.3
AidanBench	4922
Aider Polyglot	60.4
AIME 2024 (avg@8)	89.2
AIME 2024-2025	86.5
AIME 2025 (avg@8)	83.8
AIME 2025 (MathArena repeated-run mean)	86.7
Arabic	1093
ARC-AGI-1 Semi-Private	34.5
ARC-AGI-2	3.0
Arena Text — Business, Management & Financial Ops (No Style Control)	1313
Arena Text — Chinese (No Style Control)	1379
Arena Text — Creative Writing (No Style Control)	1286
Arena Text — English (No Style Control)	1351
Arena Text — Entertainment, Sports & Media (No Style Control)	1283
Arena Text — Exclude Ties (No Style Control)	1291
Arena Text — Exclude Ties (Style Controlled)	1330
Arena Text — Expert (No Style Control)	1364
Arena Text — French (No Style Control)	1334
Arena Text — French (Style Controlled)	1372
Arena Text — German (No Style Control)	1303
Arena Text — Hard Prompts (No Style Control)	1365
Arena Text — Hard Prompts English (No Style Control)	1375
Arena Text — Instruction Following (No Style Control)	1337