Which is better, o1 or o3?

Across 53 shared benchmarks tracked on Vector Wire, o1 leads on 7 and o3 leads on 43 — o3 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

o1 vs o3

OpenAIvsOpenAI53 shared benchmarks7–43 head-to-head

Benchmark	o1	o3
AA Agentic Index	31.1	36.1
AA Intelligence	24	31.1
AA-LCR	63.3	73.3
AA-Omniscience	-10.6	-15.3
aider_polyglot	61.7	81.3
AIME 2024	79.2	91.6
AIME 2025	81.7	89.2
AIR-Bench 2024	80	84.5
anthropic_red_team	98.3	98.3
arena_vision	1193	1217
Artificial Analysis Coding Index	39.7	38.4
bbq	97.3	97.9
BBQ Accuracy on Ambiguous Questions	1	0.9
BBQ Accuracy on Unambiguous Questions	0.9	0.9
BBQ P(not stereotyping \| ambiguous question, not unknown)	0.1	0.3
critpt	0.3	1.1
Fortress	19.4	16
gdpval	11.5	12.8
GPQA Diamond	75.7	83.3
harmbench	96.3	98.4
HLE	7.7	20.3
IFBench	70.3	71.4
math_500_em	96.4	98.1
MathVista	71.8	86.8
metr_hcast	42.6	64.7
metr_re_bench	0	15
metr_swaa	100	99.8
MMLU Language (0-shot) - Average	0.9	0.9
MMLU Language (0-shot) - Bengali	0.9	0.9
MMLU Language (0-shot) - Chinese (Simplified)	0.9	0.9
MMLU Language (0-shot) - French	0.9	0.9
MMLU Language (0-shot) - Hindi	0.9	0.9
MMLU Language (0-shot) - Indonesian	0.9	0.9
MMLU Language (0-shot) - Italian	0.9	0.9
MMLU Language (0-shot) - Korean	0.9	0.9
MMLU Language (0-shot) - Spanish	0.9	0.9
MMMU	77.6	82.9
OmniScience Accuracy	34.7	38.6
OmniScience Non-Hallucination	30.7	12.9
PersonQA hallucination rate	0.2	0.3
scicode	35.8	41
simple_safety_tests	99	99
simplebench	41.7	53.1
simpleqa	47	49.4
SimpleQA accuracy	0.5	0.5
SimpleQA hallucination rate	0.4	0.5
strongreject	97	97
SWE-bench Verified	48.9	69.1
TAU-bench (airline)	50	52
TAU-bench (retail)	70.8	73.9
Terminal-Bench Hard	12.9	37.1
xstest	97	97.3
τ²-Bench Telecom (AA run)	62.6	80.7

Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.