Which is better, GPT-4o or Qwen2.5 Instruct 72B?

Across 65 shared benchmarks tracked on Vector Wire, GPT-4o leads on 46 and Qwen2.5 Instruct 72B leads on 18 — GPT-4o leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

GPT-4o vs Qwen2.5 Instruct 72B

OpenAIvsAlibaba65 shared benchmarks46–18 head-to-head

Benchmark	GPT-4o	Qwen2.5 Instruct 72B
AA Intelligence	11.2	10
AA-LCR	55	20.3
AA-Omniscience	-10.5	-52.2
aider_edit_acc	72.9	65.4
aider_polyglot	30.7	7.6
AIME 2024	9.3	23.3
AIR-Bench 2024	62.4	59
anthropic_red_team	99.1	99.6
Arena Hard	92.4	81.2
Artificial Analysis Coding Index	24.2	11.9
bbq	95.1	95.4
C-Eval	76	89.2
Chinese SimpleQA (C-SimpleQA)	64.6	52.2
CLUEWSC	87.9	91.4
Codeforces	759	24.8
critpt	0	0
DROP	83.4	76.7
DROP (3-shot F1)	83.7	76.7
DROP (F1)	89.2	85
Fortress	47.2	56.4
FRAMES (Acc.)	80.5	69.8
GPQA Diamond	70.1	49.1
GSM8K	95.6	95.8
harmbench	82.9	72.8
HLE	5.3	4.2
humaneval	90.6	86.6
HumanEval-Mul (Pass@1)	80.5	77.3
IFBench	36	36.9
ifeval	84.3	87.2
IFEval (avg)	84.1	87.2
LiveCodeBench	38.3	55.5
livecodebench_pass1cot	34.2	31.1
LongBench v2 easy (w/ CoT)	54.2	47.9
LongBench v2 easy (w/o CoT)	57.4	42.7
LongBench v2 hard (w/ CoT)	49.7	40.8
LongBench v2 hard (w/o CoT)	45.6	41.8
LongBench v2 long (w/ CoT)	43.5	39.8
LongBench v2 long (w/o CoT)	40.2	44.4
LongBench v2 medium (w/ CoT)	48.6	40.9
LongBench v2 medium (w/o CoT)	52.4	38.1
LongBench v2 overall (w/ CoT)	51.4	43.5
LongBench v2 overall (w/o CoT)	50.1	42.1
LongBench v2 short (w/ CoT)	59.6	48.9
LongBench v2 short (w/o CoT)	53.3	45.6
longbench_v2	48.1	39.4
MATH	85.3	88.4
math_500_em	74.6	80
MBPP+ (EvalPlus-augmented)	76.2	77
mgsm	90.5	87.3
mmlu	88.1	86.1
mmlu_redux	88	86.8
MMLU-Pro	74.7	71.6
mmmlu	81.4	74.8
narrativeqa	80.4	74.5
naturalquestions_closedbook	50.1	35.9
OmniScience Accuracy	23.7	17.6
OmniScience Non-Hallucination	62.1	15.3
OpenBookQA	96.8	96.2
scicode	33.4	26.7
simple_safety_tests	98.5	100
simpleqa	39.4	10.3
SWE-bench Verified	38.8	23.8
Terminal-Bench Hard	8.3	4.5
xstest	97.3	97.9
τ²-Bench Telecom (AA run)	28.9	34.5

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.