Which is better, GPT-5.6 Sol or Opus 5?

Across 44 shared benchmarks tracked on Vector Wire, GPT-5.6 Sol leads on 15 and Opus 5 leads on 28 — Opus 5 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

GPT-5.6 Sol vs Opus 5

OpenAIvsAnthropic44 shared benchmarks15–28 head-to-head

Benchmark	GPT-5.6 Sol	Opus 5
AA Agentic Index	57.8	59.2
AA Intelligence	61	63.1
AA-LCR	77.7	78.7
AA-Omniscience	22	37.1
arc_agi_1	97.5	97.5
arc_agi_3	7.8	30.2
ARC-AGI-2	92.5	90.4
arena_elo	1482	1487
arena_text_factuality	1474	1481
Artificial Analysis Coding Index	78.3	78
browsecomp	90.4	90.8
coding_arena_elo	1619	1662
critpt	32.3	29.1
DeepSWE 1.1	73	68.8
Finance Agent v2	53.8	58.6
GDP (Surge AI)	30.7	85.5
gdpval	61.1	67.2
GDPval-AA v2	1748	1861
GPQA Diamond	94.6	93.7
HealthBench	57	67.1
healthbench_professional	60.5	59.8
HiL-Bench	32.3	57
HLE	49.5	64.7
HLE (with tools)	58	64.7
livebench_agentic_coding	56.2	65.2
livebench_coding	83.9	81.5
livebench_data_analysis	79.8	74.5
livebench_instruction_following	71.8	63.8
livebench_language	87.7	88.7
livebench_math	96.2	95.7
livebench_reasoning	91.7	91.2
MCP Atlas	83.6	85.8
MMMU-Pro	84.6	84.7
officeqa_pro	63.2	66.9
OmniScience Accuracy	59.4	60.9
OmniScience Non-Hallucination	10.6	40.5
OSWorld 2.0	62.6	70.6
OSWorld-Verified	83	4
scicode	56.9	55.7
simplebench	64.8	80.6
simpleqa_verified	71.6	56.7
SWE-bench Pro	64.6	79.2
TauBench V3 - Banking	44.3	44.7
Terminal-Bench 2.1	89.5	89.1

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.