Which is better, DeepSeek-V4-Flash or GPT-5.5?

Across 48 shared benchmarks tracked on Vector Wire, DeepSeek-V4-Flash leads on 8 and GPT-5.5 leads on 40 — GPT-5.5 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

DeepSeek-V4-Flash vs GPT-5.5

DeepSeekvsOpenAI48 shared benchmarks8–40 head-to-head

Benchmark	DeepSeek-V4-Flash	GPT-5.5
AA Agentic Index	61.3	47.4
AA Intelligence	51.8	56.3
AA-LCR	74.3	79
AA-Omniscience	-14.3	20.5
Agents' Last Exam	25.2	26.6
aime_2026	95.8	100
arc_agi_1	84	95
ARC-AGI-2	46	85
arena_elo	1439	1482
arena_text_factuality	1434	1482
Artificial Analysis Coding Index	69.1	74.9
browsecomp	73.2	84.4
coding_arena_elo	1431	1457
critpt	16.6	27.1
cybergym	76.7	81.8
DeepSWE	54.4	70
gdpval	52.9	49.5
GDPval-AA (Elo)	1395	1769
GDPval-AA v2	1189	1509
GPQA Diamond	90.8	93.6
HLE	38.6	52.2
HLE (with tools)	45.1	52.2
hmmt_feb_2026	94.8	98.5
IFBench	79.2	75.9
itbenchSre	31.5	45.8
livebench_agentic_coding	46.8	54
livebench_coding	75	82.2
livebench_data_analysis	79.3	81.6
livebench_instruction_following	65.5	70.7
livebench_language	79.2	87.4
livebench_math	86.8	95.9
livebench_reasoning	86.6	89.7
MCP Atlas	69	82.8
nl2repo	54.2	50.7
OmniScience Accuracy	40.4	58
OmniScience Non-Hallucination	15.6	12
scicode	49.9	56.1
simplebench	46.3	69
simpleqa_verified	45	63.1
SWE-bench Pro	52.6	58.6
SWE-bench Verified	79	80.6
TauBench V3 - Banking	39.4	39
Terminal-Bench 2.0	56.9	82.7
Terminal-Bench 2.1	82.7	84.3
Terminal-Bench Hard	38.6	60.6
toolathlon	70.3	55.6
τ²-Bench Telecom (AA run)	95.6	93.9
τ³-Bench	22.9	31.3

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.