OpenAI o3 vs Claude Opus 5

New Comparison

Scores are fetched live from the current database on every visit — nothing is cached.

Benchmark Scores Comparison

OpenAI o3 Claude Opus 5

OpenAI o3

Score: 15.0

Claude Opus 5

Score: 40.0

Coverage matrix

2 models × 55 benchmarks — hover a benchmark name for its category

Full Coverage Matrix
Hides benchmarks where all selected models score identically.
Shading = score relative to the best score among the selected models: Low Mid High Missing (—) The totals row shows how many of the 55 benchmarks each model has scores for.
Benchmark OpenAI o3 OpenAI Claude Opus 5 Anthropic
GPQA Diamond — 93.7
ArXivMath — 71.5
Hy-LifeSearch (Internal) — 56.1
Automation-Bench — 48.7
SWE-bench Pro — 79.2
BioMysteryBench — 72.1
Agents' Last Exam — 25.1
Toolathlon Verified — 76.5
Hy-Backend 2.0 (Internal) — 40.3
Hy-SWE Max Verified (Internal) — 70.1
Hy-BrowseComp-Pro2 (Internal) — 61.3
OfficeQA Pro — 66.9
SWE Atlas - TW — 68.7
LOCA-Bench (256k) — 53.7
SWE Atlas - QnA — 57.5
NL2Repo — 75.3
PostTrainBench — 35
BabyVision — 94.1
MMMLU 88.8 —
Apex-Agents — 41.8
WorkSpaceBench — 75
BankerToolBench — 81.9
E-Bench-Code (Internal) — 82.7
HLE (with tools) — 60.9
CharXiv (RQ) — 83.7
GDP.pdf — 21.6
CritPt (no tools) — 29.1
NL2Repo-Bench — 75.3
DRACO — 88.6
SWE-bench Multilingual — 89.5
Hy-FinAgentBench (Internal) — 82
Humanity's Last Exam — 54.9
SUPERChem — 76.7
HorizonMath (pass@4) — 5.3
Terminal-Bench 2.1 (Terminus-2) — 86.7
DeepSWE — 68.8
BrokenArXiv — 77.7
Hy-FinmodelBench v2 (Internal) — 66
MathArena Apex 2025 — 91.4
JobBench — 68
$OneMillion-Bench — 68.1
OSWorld 2.0 (Binary) — 70.6
WideSearch — 84
SkillsBench Avg5 — 63.7
E-Bench (Internal) — 77.8
ProgramBench — 39.5
MCP-Atlas — 85.7
SWE-Marathon — 50
Terminal-Bench 4.0 — 51.8
SWE Atlas - RF — 60
DeepSWE 1.1 — 74
Hy-CompanyBench V2 (Internal) — 72.7
Harbor-Index — 56.9
SWE Refactor Bench — 31
GDPVal-AA v2 — 1831
Covered 1/55 54/55

Scorecard

Overall 0–100 score — category strengths (50%), shared-benchmark wins (30%), coverage (20%)

Model Overall score Shared benchmark wins Coverage Strongest at Weakest at
Claude Opus 5 98.2 — 54 of 55 (98.0%) — —
OpenAI o3 1.8 — 1 of 55 (2.0%) — —

Partial coverage benchmarks

Scored by at least one but not all selected models (55) — missing scores show as —

Benchmark Category OpenAI o3 Claude Opus 5
GPQA Diamond 1/2 models stem_reasoning — 93.7
ArXivMath 1/2 models stem_reasoning — 71.5
Hy-LifeSearch (Internal) 1/2 models general_agent — 56.1
Automation-Bench 1/2 models general_agent — 48.7
SWE-bench Pro 1/2 models coding_agent — 79.2
BioMysteryBench 1/2 models stem_reasoning — 72.1
Agents' Last Exam 1/2 models general_agent — 25.1
Toolathlon Verified 1/2 models general_agent — 76.5
Hy-Backend 2.0 (Internal) 1/2 models coding_agent — 40.3
Hy-SWE Max Verified (Internal) 1/2 models coding_agent — 70.1
Hy-BrowseComp-Pro2 (Internal) 1/2 models general_agent — 61.3
OfficeQA Pro 1/2 models general_agent — 66.9
SWE Atlas - TW 1/2 models coding_agent — 68.7
LOCA-Bench (256k) 1/2 models long_context — 53.7
SWE Atlas - QnA 1/2 models coding_agent — 57.5
NL2Repo 1/2 models coding_agent — 75.3
PostTrainBench 1/2 models coding_agent — 35
BabyVision 1/2 models vision_language — 94.1
MMMLU 1/2 models multilingual 88.8 —
Apex-Agents 1/2 models general_agent — 41.8
WorkSpaceBench 1/2 models general_agent — 75
BankerToolBench 1/2 models general_agent — 81.9
E-Bench-Code (Internal) 1/2 models coding_agent — 82.7
HLE (with tools) 1/2 models stem_reasoning — 60.9
CharXiv (RQ) 1/2 models document_understanding — 83.7
GDP.pdf 1/2 models long_context — 21.6
CritPt (no tools) 1/2 models stem_reasoning — 29.1
NL2Repo-Bench 1/2 models coding_agent — 75.3
DRACO 1/2 models general_agent — 88.6
SWE-bench Multilingual 1/2 models coding_agent — 89.5
Hy-FinAgentBench (Internal) 1/2 models domain_finance — 82
Humanity's Last Exam 1/2 models stem_reasoning — 54.9
SUPERChem 1/2 models stem_reasoning — 76.7
HorizonMath (pass@4) 1/2 models stem_reasoning — 5.3
Terminal-Bench 2.1 (Terminus-2) 1/2 models coding_agent — 86.7
DeepSWE 1/2 models coding_agent — 68.8
BrokenArXiv 1/2 models stem_reasoning — 77.7
Hy-FinmodelBench v2 (Internal) 1/2 models domain_finance — 66
MathArena Apex 2025 1/2 models stem_reasoning — 91.4
JobBench 1/2 models general_agent — 68
$OneMillion-Bench 1/2 models general_capabilities — 68.1
OSWorld 2.0 (Binary) 1/2 models general_agent — 70.6
WideSearch 1/2 models general_agent — 84
SkillsBench Avg5 1/2 models coding_agent — 63.7
E-Bench (Internal) 1/2 models general_agent — 77.8
ProgramBench 1/2 models coding_agent — 39.5
MCP-Atlas 1/2 models general_agent — 85.7
SWE-Marathon 1/2 models coding_agent — 50
Terminal-Bench 4.0 1/2 models coding_agent — 51.8
SWE Atlas - RF 1/2 models coding_agent — 60
DeepSWE 1.1 1/2 models coding_agent — 74
Hy-CompanyBench V2 (Internal) 1/2 models general_agent — 72.7
Harbor-Index 1/2 models coding_agent — 56.9
SWE Refactor Bench 1/2 models coding_agent — 31
GDPVal-AA v2 1/2 models general_agent — 1831

What now?

You've seen 2 models side by side — keep the loop going.

Share this comparison

The URL is the comparison — copy or share it anywhere. No tracking, no cookies.

Not the right matchup?

Swap one model out — the pre-filled form keeps the rest of the selection.

Swap OpenAI o3 × Swap Claude Opus 5 ×