Claude Opus 4.6 vs DeepSeek V4 Pro vs Edge0-145B-NVFP4 vs NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB

New Comparison

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

Winner

DeepSeek V4 Pro

DeepSeek V4 Pro scores higher overall across category strength profiles (92 vs 48 for Claude Opus 4.6); only 1 benchmarks are directly comparable.

DeepSeek V4 Pro 91.7 Claude Opus 4.6 47.5 Edge0-145B-NVFP4 33.8 NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB 28.7

Only 1 of 87 benchmarks are scored by every selected model — the verdict leans on category strength profiles and is indicative.

Benchmark Scores Comparison

Claude Opus 4.6 DeepSeek V4 Pro Edge0-145B-NVFP4 NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB

Claude Opus 4.6

Score: 15.0

DeepSeek V4 Pro

Score: 15.0

Edge0-145B-NVFP4

145.0B params

NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB

74.0B params

Coverage matrix

4 models × 87 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 87 benchmarks each model has scores for.
Benchmark NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB NVIDIA Edge0-145B-NVFP4 Edge0 Claude Opus 4.6 Anthropic DeepSeek V4 Pro DeepSeek
OmniScience Accuracy 16.9 — — —
SVG-Bench — — — 61.7
GPQA Diamond 86.43 90.4 91.3 92.8
ArXivMath — — — 62.1
LiveCodeBench v6 — — 88.8 —
Hy-LifeSearch (Internal) — — — 46.9
Automation-Bench — — — 30.4
SWE-bench Pro — — 53.4 55.4
BioMysteryBench — — — 61.6
AA-LCR 64.06 79.3 — —
MathVision — — 71.2 —
AIME 26 — — 96.7 —
AgentWorldBench Terminal — — 57.51 —
Agents' Last Exam — — — 21.9
Toolathlon Verified — — 47.2 74.1
Hy-Backend 2.0 (Internal) — — — 33.8
OJBench — — 60.3 —
Hy-SWE Max Verified (Internal) — — — 65.7
AgentWorldBench Overall — — 57.8 —
OmniScience Non-Hallucination 61.27 — — —
Hy-BrowseComp-Pro2 (Internal) — — — 46.5
GDPVal — — — 70.3
OfficeQA Pro — — — 65.4
DeepSearch QA — — 91.3 —
SWE Atlas - TW — — — 45.6
AgentWorldBench Web — — 51.42 —
SWE Atlas - QnA — — — 53.4
NL2Repo — — — 61.5
MMMU-Pro — — 73.9 —
PostTrainBench — — — 24.5
BabyVision — — 14.8 —
Claw-Eval Avg — — 82.4 58.4
Apex-Agents — — 33 32.4
PinchBench 87.82 — — —
WorkSpaceBench — — — 65.4
BankerToolBench — — — 73.1
E-Bench-Code (Internal) — — — 64.3
OSWorld-Verified — — 72.7 —
AgentWorldBench Search — — 29.3 —
HLE (with tools) — 35.4 53 60
CharXiv (RQ) — — 69.1 —
CritPt (no tools) — — — 18
BrowseComp 66.75 — 83.7 83.4
NL2Repo-Bench — — — 35.5
DRACO — — — 77.3
MCPMark — — 56.7 —
SWE-bench Multilingual — — 77.8 77.3
YC-Bench — — — 500000
Hy-FinAgentBench (Internal) — — — 78.5
IFBench — 80.6 — —
SWE-bench Verified — — 80.8 80.6
IMOAnswerBench — — 75.3 —
Humanity's Last Exam 25.3 — 40 42.7
Terminal-Bench 2.0 — — 65.4 —
SpreadSheetBench-v1 — — — 84.9
SUPERChem — — — 62
VIBE-V2 — — — 45
HorizonMath (pass@4) — — — 4.42
Terminal-Bench 2.1 (Terminus-2) — 86.5 — 87.9
DeepSWE — — — 62.7
BrokenArXiv — — — 43.1
HMMT Feb 26 — — 96.2 —
AgentWorldBench OS — — 70.2 —
AgentWorldBench MCP — — 69.9 —
Hy-FinmodelBench v2 (Internal) — — — 51.3
MathArena Apex 2025 — — — 66.3
Terminal Bench 2.1 50.98 — — 59.6
AgentWorldBench SWE — — 64.55 —
JobBench — — — 54.1
$OneMillion-Bench — — — 62
SciCode (subtask) — — 51.9 —
WideSearch — — — 81.8
SkillsBench Avg5 — — — 65
Claw-Eval Pass^3 — — 70.4 —
E-Bench (Internal) — — — 61.3
ProgramBench — — — 15.5
MCP-Atlas 52.2 — — 73.6
SWE-Marathon — — — 19
Cybergym — — — 83.3
SWE Atlas - RF — — — 48.6
V-Star — — 86.4 —
AgentWorldBench Android — — 61.74 —
DeepSWE 1.1 — 64.6 — —
Hy-CompanyBench V2 (Internal) — — — 64.4
Harbor-Index — — — 36.9
SciCode 41.02 — — —
GDPVal-AA v2 28.47 — — 1580
Covered 11/87 6/87 34/87 56/87

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
DeepSeek V4 Pro Winner 91.7 1 of 1 56 of 87 (64.0%) STEM & Reasoning Agentic Coding
Claude Opus 4.6 47.5 0 of 1 34 of 87 (39.0%) STEM & Reasoning General Agents
Edge0-145B-NVFP4 33.8 0 of 1 6 of 87 (7.0%) Long Context STEM & Reasoning
NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB 28.7 0 of 1 11 of 87 (13.0%) Long Context General Agents

Category strengths

Where each model stands, per benchmark category — score relative to the best model, adjusted for benchmark coverage. This also ranks models when they share few benchmarks.

Category DeepSeek V4 Pro Claude Opus 4.6 Edge0-145B-NVFP4 NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB
General Agents 33 benchmarks 99.5 66.1 — 39.8
Agentic Coding 25 benchmarks 94.8 84.2 61.2 53.4
STEM & Reasoning 16 benchmarks 100 91.7 60.3 61.5
Long Context 1 benchmark — — 100 80.8

Fully comparable benchmarks

Scored by all 4 selected models (1)

Benchmark Category Claude Opus 4.6 DeepSeek V4 Pro Edge0-145B-NVFP4 NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB
GPQA Diamond stem_reasoning 91.3 92.8 90.4 86.43

Partial coverage benchmarks

Scored by at least one but not all selected models (86) — missing scores show as —

Benchmark Category Claude Opus 4.6 DeepSeek V4 Pro Edge0-145B-NVFP4 NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB
OmniScience Accuracy 1/4 models knowledge — — — 16.9
SVG-Bench 1/4 models coding_agent — 61.7 — —
ArXivMath 1/4 models stem_reasoning — 62.1 — —
LiveCodeBench v6 1/4 models stem_reasoning 88.8 — — —
Hy-LifeSearch (Internal) 1/4 models general_agent — 46.9 — —
Automation-Bench 1/4 models general_agent — 30.4 — —
SWE-bench Pro 2/4 models coding_agent 53.4 55.4 — —
BioMysteryBench 1/4 models stem_reasoning — 61.6 — —
AA-LCR 2/4 models long_context — — 79.3 64.06
MathVision 1/4 models vision_language 71.2 — — —
AIME 26 1/4 models stem_reasoning 96.7 — — —
AgentWorldBench Terminal 1/4 models general_agent 57.51 — — —
Agents' Last Exam 1/4 models general_agent — 21.9 — —
Toolathlon Verified 2/4 models general_agent 47.2 74.1 — —
Hy-Backend 2.0 (Internal) 1/4 models coding_agent — 33.8 — —
OJBench 1/4 models stem_reasoning 60.3 — — —
Hy-SWE Max Verified (Internal) 1/4 models coding_agent — 65.7 — —
AgentWorldBench Overall 1/4 models general_agent 57.8 — — —
OmniScience Non-Hallucination 1/4 models knowledge — — — 61.27
Hy-BrowseComp-Pro2 (Internal) 1/4 models general_agent — 46.5 — —
GDPVal 1/4 models general_agent — 70.3 — —
OfficeQA Pro 1/4 models general_agent — 65.4 — —
DeepSearch QA 1/4 models general_agent 91.3 — — —
SWE Atlas - TW 1/4 models coding_agent — 45.6 — —
AgentWorldBench Web 1/4 models general_agent 51.42 — — —
SWE Atlas - QnA 1/4 models coding_agent — 53.4 — —
NL2Repo 1/4 models coding_agent — 61.5 — —
MMMU-Pro 1/4 models vision_language 73.9 — — —
PostTrainBench 1/4 models coding_agent — 24.5 — —
BabyVision 1/4 models vision_language 14.8 — — —
Claw-Eval Avg 2/4 models coding_agent 82.4 58.4 — —
Apex-Agents 2/4 models general_agent 33 32.4 — —
PinchBench 1/4 models general_agent — — — 87.82
WorkSpaceBench 1/4 models general_agent — 65.4 — —
BankerToolBench 1/4 models general_agent — 73.1 — —
E-Bench-Code (Internal) 1/4 models coding_agent — 64.3 — —
OSWorld-Verified 1/4 models general_agent 72.7 — — —
AgentWorldBench Search 1/4 models general_agent 29.3 — — —
HLE (with tools) 3/4 models stem_reasoning 53 60 35.4 —
CharXiv (RQ) 1/4 models document_understanding 69.1 — — —
CritPt (no tools) 1/4 models stem_reasoning — 18 — —
BrowseComp 3/4 models general_agent 83.7 83.4 — 66.75
NL2Repo-Bench 1/4 models coding_agent — 35.5 — —
DRACO 1/4 models general_agent — 77.3 — —
MCPMark 1/4 models general_agent 56.7 — — —
SWE-bench Multilingual 2/4 models coding_agent 77.8 77.3 — —
YC-Bench 1/4 models general_agent — 500000 — —
Hy-FinAgentBench (Internal) 1/4 models domain_finance — 78.5 — —
IFBench 1/4 models instruction_following — — 80.6 —
SWE-bench Verified 2/4 models coding_agent 80.8 80.6 — —
IMOAnswerBench 1/4 models stem_reasoning 75.3 — — —
Humanity's Last Exam 3/4 models stem_reasoning 40 42.7 — 25.3
Terminal-Bench 2.0 1/4 models coding_agent 65.4 — — —
SpreadSheetBench-v1 1/4 models general_agent — 84.9 — —
SUPERChem 1/4 models stem_reasoning — 62 — —
VIBE-V2 1/4 models coding_agent — 45 — —
HorizonMath (pass@4) 1/4 models stem_reasoning — 4.42 — —
Terminal-Bench 2.1 (Terminus-2) 2/4 models coding_agent — 87.9 86.5 —
DeepSWE 1/4 models coding_agent — 62.7 — —
BrokenArXiv 1/4 models stem_reasoning — 43.1 — —
HMMT Feb 26 1/4 models stem_reasoning 96.2 — — —
AgentWorldBench OS 1/4 models general_agent 70.2 — — —
AgentWorldBench MCP 1/4 models general_agent 69.9 — — —
Hy-FinmodelBench v2 (Internal) 1/4 models domain_finance — 51.3 — —
MathArena Apex 2025 1/4 models stem_reasoning — 66.3 — —
Terminal Bench 2.1 2/4 models coding_agent — 59.6 — 50.98
AgentWorldBench SWE 1/4 models general_agent 64.55 — — —
JobBench 1/4 models general_agent — 54.1 — —
$OneMillion-Bench 1/4 models general_capabilities — 62 — —
SciCode (subtask) 1/4 models stem_reasoning 51.9 — — —
WideSearch 1/4 models general_agent — 81.8 — —
SkillsBench Avg5 1/4 models coding_agent — 65 — —
Claw-Eval Pass^3 1/4 models coding_agent 70.4 — — —
E-Bench (Internal) 1/4 models general_agent — 61.3 — —
ProgramBench 1/4 models coding_agent — 15.5 — —
MCP-Atlas 2/4 models general_agent — 73.6 — 52.2
SWE-Marathon 1/4 models coding_agent — 19 — —
Cybergym 1/4 models general_agent — 83.3 — —
SWE Atlas - RF 1/4 models coding_agent — 48.6 — —
V-Star 1/4 models vision_language 86.4 — — —
AgentWorldBench Android 1/4 models general_agent 61.74 — — —
DeepSWE 1.1 1/4 models coding_agent — — 64.6 —
Hy-CompanyBench V2 (Internal) 1/4 models general_agent — 64.4 — —
Harbor-Index 1/4 models coding_agent — 36.9 — —
SciCode 1/4 models — — — — 41.02
GDPVal-AA v2 2/4 models general_agent — 1580 — 28.47

Comparison FAQ

Which model scores highest in this comparison?

DeepSeek V4 Pro scores higher overall across category strength profiles (92 vs 48 for Claude Opus 4.6); only 1 benchmarks are directly comparable.

Which model is strongest at Agentic Coding?

DeepSeek V4 Pro leads the Agentic Coding category (strength score 94.8/100 across 6 comparable benchmarks).

How reliable is this comparison verdict?

Only 1 of 87 benchmarks are scored by every selected model — the verdict leans on category strength profiles and is indicative.

What now?

You've seen 4 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?

Open the create form with these models pre-filled and untick the one you want to replace.

Swap a model