Which is better, DeepSeek-V3.2 or GPT-5?

Across 77 shared benchmarks tracked on Vector Wire, DeepSeek-V3.2 leads on 21 and GPT-5 leads on 55 — GPT-5 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

DeepSeek-V3.2 vs GPT-5

DeepSeekvsOpenAI77 shared benchmarks21–55 head-to-head

Benchmark	DeepSeek-V3.2	GPT-5
AA Agentic Index	39.8	49.7
AA Intelligence	32.8	35.3
AA-LCR	70.7	76.3
AA-Omniscience	-22.5	-8.7
AIME 2025	93.1	95
AIME25 no tools	89.3	94.6
AIME25 w/ python	58.1	99.6
arc_agi_1	57	56.2
ARC-AGI-2	4	7.5
Arena-Hard (Creative Writing)	88.8	92.2
Arena-Hard (Hard Prompt)	53.4	71.9
ArtifactsBench	55.8	73
Artificial Analysis Coding Index	44.2	38.9
browsecomp	51.4	54.9
BrowseComp w/ tools	40.1	54.9
browsecomp_zh	65	65
BrowseComp-ZH w/ tools	47.9	63
coding_arena_elo	1368	1419
critpt	2.9	5.7
FinSearchComp-global	26.2	63.9
FinSearchComp-T3	27	48.5
FinSearchComp-T3 w/ tools	27	48.5
Frames	80.2	86
Frames w/ tools	80.2	86
frontiermath_tier_4	2.1	12.5
GAIA (text only)	63.5	76.4
gaia_level1	90.3	78.5
gaia_level2	73.6	64.8
gaia_level3	61.2	46.9
gaia_overall	76.7	66.1
gdpval	18.8	32.5
GPQA (unspecified)	79.9	85.7
GPQA Diamond	84	87.3
HealthBench	46.9	67.2
HealthBench no tools	46.9	67.2
HLE	40.8	28.5
HLE (with tools)	40.8	35.2
HMMT 2025	90.2	93.3
hmmt_feb_2025	92.5	88.3
hmmt_nov_2025	90.2	89.2
HMMT25 no tools	83.6	93.3
HMMT25 w/ python	49.5	96.7
IFBench	61	73.1
imo_answer_bench	78.3	76
LiveCodeBench	86	85
LiveCodeBench v6	83.3	87
LiveCodeBenchV6 no tools	74.1	87
Longform Writing eval (Kimi K2 Thinking system card)	72.5	71.4
mmlu_redux	93.7	95.3
MMLU-Pro	86	87.5
Multi-SWE-Bench	37.4	39.3
OJ-Bench (cpp)	54.7	56.2
OJ-Bench (cpp) no tools	38.2	56.2
OmniScience Accuracy	33	40.3
OmniScience Non-Hallucination	17.3	21
scicode	39	43
Seal-0	49.5	51.4
Seal-0 w/ tools	38.5	51.4
simpleqa_verified	27.5	50.6
swe_bench_bash	60	65
SWE-bench Multilingual	70.2	55.3
SWE-bench Pro	15.6	41.8
SWE-bench Verified	73.1	74.9
TauBench V3 - Banking	18.8	22.1
Terminal-Bench	37.7	43.8
Terminal-Bench 2.0	46.4	35.2
Terminal-Bench 2.1	46.8	35.2
Terminal-Bench Hard	35.6	37.9
Terminal-Bench w/ simulated tools (JSON)	37.7	43.8
vectara_answer_rate	92.6	99.9
vectara_avg_summary_length	62	162.7
vectara_factual_consistency	93.7	85.3
vectara_hallucination_rate ↓	6.3	14.7
xbench_deepsearch	71	77.8
τ²-Bench	91	85
τ²-Bench Telecom (AA run)	90.6	86.5
τ³-Bench	69.2	19.6

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.