Which is better, DeepSeek-V4-Pro or GPT-5.6 Sol?

Across 49 shared benchmarks tracked on Vector Wire, DeepSeek-V4-Pro leads on 6 and GPT-5.6 Sol leads on 43 — GPT-5.6 Sol 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-Pro vs GPT-5.6 Sol

DeepSeekvsOpenAI49 shared benchmarks6–43 head-to-head

Benchmark	DeepSeek-V4-Pro	GPT-5.6 Sol
AA Agentic Index	63.3	57.8
AA Intelligence	53	61
AA-LCR	70	77.7
AA-Omniscience	-10.6	22
Agents' Last Exam	16.5	52.7
aime_2026	96.7	99.9
arena_elo	1457	1482
arena_text_factuality	1450	1474
Artificial Analysis Coding Index	59.4	78.3
browsecomp	83.4	90.4
BrowseComp (w/ Ctx)	83.4	89.4
coding_arena_elo	1582	1619
critpt	13	32.3
cybergym	52.7	83.6
DeepSWE	12.8	73
FORTRESS (Adversarial)	36	82.4
FORTRESS (Benign)	98.5	98.1
gdpval	49	61.1
GDPval-AA v2	1307	1748
Global-MMLU-Lite	89.3	91.8
GPQA Diamond	90.5	94.6
HLE	48.2	49.5
HLE (with tools)	48.2	58
IFBench	76.5	72.7
itbenchSre	38.3	56.2
livebench_agentic_coding	42.6	56.2
livebench_coding	70	83.9
livebench_data_analysis	74.5	79.8
livebench_instruction_following	62.4	71.8
livebench_language	78.1	87.7
livebench_math	90.7	96.2
livebench_reasoning	82.7	91.7
MCP Atlas	74.2	83.6
OmniScience Accuracy	43	59.4
OmniScience Non-Hallucination	12.2	10.6
Program Bench	47.8	77.6
scicode	50	56.9
simplebench	50.9	64.8
simpleqa_verified	57.9	71.6
strongreject	98.6	98.5
SWE-bench Pro	55.4	64.6
SWEBench Pro (Public)	55.4	64.6
Tau 3 Banking	26	33
TauBench V3 - Banking	30.1	44.3
Terminal Bench 2.1 (Best Harness)	64	89.5
Terminal-Bench 2.1	72.1	89.5
Terminal-Bench Hard	46.2	65.9
toolathlon	51.8	58
τ²-Bench Telecom (AA run)	96.2	85.1

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.