Which is better, Claude Opus 4.8 or GPT-5.6 Sol?

Across 79 shared benchmarks tracked on Vector Wire, Claude Opus 4.8 leads on 20 and GPT-5.6 Sol leads on 56 — 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 —

Claude Opus 4.8 vs GPT-5.6 Sol

AnthropicvsOpenAI79 shared benchmarks20–56 head-to-head

Benchmark	Claude Opus 4.8	GPT-5.6 Sol
AA Agentic Index	49.4	57.8
AA Intelligence	57.3	61
AA-LCR	73	77.7
AA-Omniscience	28.8	22
Agents' Last Exam	27	52.7
Agents' Last Exam (Pass / Score)	45.1	53.6
aime_2026	100	99.9
AndroidBench	69.8	74
apex_agents	39.4	56.7
arc_agi_1	92.5	97.5
arc_agi_3	1.5	7.8
ARC-AGI-2	72.1	92.5
arena_elo	1482	1482
arena_text_factuality	1464	1474
Artificial Analysis Coding Index	74.3	78.3
baby_vision_with_python	81.2	88.9
browsecomp	88.5	90.4
charxiv_rq	80.5	87.8
coding_arena_elo	1539	1619
CorpFin v2	66.7	64.4
critpt	20.9	32.3
cybergym	83.1	83.6
DeepSWE	59	73
DeepSWE 1.1	59	73
Finance Agent v2	53.9	53.8
GDP (Surge AI)	84.8	30.7
gdpval	54.2	61.1
GDPval-AA v2	1593	1748
GDPval-AA v2 (Elo)	1593	1736
GPQA Diamond	93.6	94.6
HealthBench	59.3	57
healthbench_professional	57.4	60.5
HiL-Bench	35.3	32.3
HLE	57.9	49.5
HLE (with tools)	57.9	58
IFBench	62.2	72.7
JobBench	48.4	45.4
Legal Research Bench	43.8	48.1
livebench_agentic_coding	50.5	56.2
livebench_coding	81.8	83.9
livebench_data_analysis	66	79.8
livebench_instruction_following	72	71.8
livebench_language	79.7	87.7
livebench_math	94.3	96.2
livebench_reasoning	89.2	91.7
longbench_v2	69.1	67.1
mathvision	86.7	95.8
MCP Atlas	83.6	83.6
MCP Mark Verified	76.4	92.9
MLS Bench Lite	42.8	46.2
MMMU-Pro	78.9	84.6
MMVU	79.2	81.2
MRCR v2 256K (8-needle)	83.2	93.8
officeqa_pro	66.2	63.2
OmniDocBench	87.9	85.8
OmniScience Accuracy	48.8	59.4
OmniScience Non-Hallucination	60.7	10.6
OSWorld 2.0	55.7	62.6
OSWorld-Verified	83.4	83
PaperBench	80.3	90.5
PerceptionBench	47.2	59.7
PostTrainBench	37.2	34.6
Program Bench	71.9	77.6
ResearchRubrics	73.5	73.8
SaaS-Bench	56.1	61.4
scicode	53.5	56.9
simplebench	64.8	64.8
simpleqa_verified	39.5	71.6
SpreadsheetBench 2	31.6	32.4
SWE-bench Pro	69.2	64.6
SWE-Marathon	40	39
Tau 3 Banking	27.6	33
TauBench V3 - Banking	34.2	44.3
Terminal-Bench 2.1	85	89.5
Terminal-Bench Hard	58.3	65.9
toolathlon	59.9	58
Video-MME	86	89.5
WorldVQA ForceAnswer	39.1	41.8
τ²-Bench Telecom (AA run)	94.4	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.