DeepSeek V4 Pro vs Qwen3.8-Flash-Next vs Claude Opus 5 vs MiniMax-M3.1-Flash vs Hy4 flash vs Ripple-65B Preview

New Comparison

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

Winner

Claude Opus 5

Claude Opus 5 scores higher overall across category strength profiles (83 vs 57 for DeepSeek V4 Pro); only 0 benchmarks are directly comparable.

Claude Opus 5 82.8 DeepSeek V4 Pro 57.4 Hy4 flash 44.4 MiniMax-M3.1-Flash 43.7 Qwen3.8-Flash-Next 42.6 Ripple-65B Preview 41.4

Only 0 of 135 benchmarks are scored by every selected model — the verdict leans on category strength profiles and is indicative.

Benchmark Scores Comparison

DeepSeek V4 Pro Qwen3.8-Flash-Next Claude Opus 5 MiniMax-M3.1-Flash Hy4 flash Ripple-65B Preview

DeepSeek V4 Pro

Score: 15.0

Qwen3.8-Flash-Next

125.0B params

Score: 70.0

Claude Opus 5

Score: 40.0

MiniMax-M3.1-Flash

428.0B params

Hy4 flash

220.0B params

Ripple-65B Preview

65.0B params

Coverage matrix

6 models × 135 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 135 benchmarks each model has scores for.
Benchmark Qwen3.8-Flash-Next Qwen Team (Alibaba) MiniMax-M3.1-Flash MiniMaxAI Ripple-65B Preview Ricursive Intelligence Hy4 flash Tencent DeepSeek V4 Pro DeepSeek Claude Opus 5 Anthropic
AGIEval (Chinese) — — — 90.6 — —
SVG-Bench — — — — 61.7 —
OSWorld-G — — — — — 76.8
GPQA Diamond 91.7 — 85.2 — 92.8 93.7
CountBenchQA — — — 86.6 — —
OCW Courses — — — 64 — —
ArXivMath — — — — 62.1 71.5
Chinese-SimpleQA — — — 74.6 — —
LiveCodeBench v6 91.9 — — — — —
Hy-LifeSearch (Internal) — — — — 46.9 56.1
Automation-Bench — 62.8 — — 30.4 48.7
SWE-bench Pro 62.5 — 61.42 — 55.4 79.2
BFCL-V4 — — — — — 80.6
RealWorldQA 88.5 — — 71.1 — —
Agents' Last Exam (Score) 51.2 — — — — —
BioMysteryBench — — — — 61.6 72.1
AA-LCR 79.1 — 75 — — —
MathVision 90.6 — — — — —
AIME 26 — — 95 — — —
AndroidWorld 84.5 — — — — —
Agents' Last Exam 24.3 29.6 — — 21.9 25.1
Toolathlon Verified 73.5 — — — 74.1 76.5
Hy-Backend 2.0 (Internal) — — — — 33.8 40.3
RecreationBench 49.9 — — — — —
Hy-SWE Max Verified (Internal) — — — — 65.7 70.1
Hy-BrowseComp-Pro2 (Internal) — — — — 46.5 61.3
ARC-c — — — 95 — —
AutomationBench (Public) — — 36.4 — — —
GDPVal — — — — 70.3 80
OfficeQA Pro — — — — 65.4 66.9
PIQA — — — 84.4 — —
StrongREJECT v2 (adaptive) — — 97.5 — — —
SWE Atlas - TW — — — — 45.6 68.7
MLE-bench Lite — — — — — 88
MBPP+ — — — 73.7 — —
LOCA-Bench (256k) — 60.3 — — — 53.7
WorkBuddy Blind User Evaluation (satisfaction) — — — 2.79 — —
SWE-bench Pro v2.0 Hard — — 79.1 — — —
SWE Atlas - QnA — — — — 53.4 57.5
ApexBench — — — — — 65.8
TextVQA — — — 73.4 — —
NL2Repo — — — — 61.5 75.3
MLS-Bench — — — — — 49.8
MMMU-Pro — — — 52 — —
Chartography — — — — — 84
PostTrainBench — 39.3 — — 24.5 35
INCLUDE — — — 82.8 — —
BabyVision — 91 — — — 94.1
Claw-Eval Avg 60.4 — — — 58.4 —
OSWorld V2.1 — — — — — 74
Apex-Agents — — — — 32.4 41.8
WorkSpaceBench — — — — 65.4 75
MMLU-Pro — — 84.2 71.1 — —
Terminal-Bench Science 0.1 — 3.3 — — — —
BankerToolBench — 76.2 — — 73.1 81.9
SuperGPQA — — — 57.8 — —
E-Bench-Code (Internal) — — — — 64.3 82.7
OSWorld-Verified — — — — — 86.4
CV-Bench — — — 83.4 — —
DeepresearchBII — — — — — 54.1
Workspace-Bench-Lite — — — — — 68.1
SpreadsheetBench 2 — — — — — 33.6
HellaSwag — — — 85.2 — —
HLE (with tools) 35.9 — — — 60 60.9
LVBench 76.6 — — — — —
CharXiv (RQ) 84.6 82.5 — — — 83.7
WorkIndex (Tencent internal) — — — 42.1 — —
GDP.pdf — 20.8 — — — 21.6
AIME 2024 — — 97.1 — — —
Belebele (5-language subset) — — — 93.8 — —
KOR-Bench — — — 54.7 — —
CMMLU — — — 91.6 — —
CritPt (no tools) — — — — 18 29.1
BrowseComp — — — — 83.4 90.8
MMLU-Redux — — — 87.6 — —
NL2Repo-Bench 48.1 61.2 — — 35.5 75.3
DRACO — — — — 77.3 88.6
SkillsBench — — — — — 63.7
WebArena-Verified — — — — — 69.1
CoWorkBench 73.9 — — — — —
SWE-bench Multilingual 81 — — — 77.3 89.5
MMLU — — 89.6 88.7 — —
Harvey Legal Agent — — — — — 22.7
PaperBench — — — — — 86.9
YC-Bench — — — — 500000 —
Agents' Last Exam (CLI) — — — — — 29.5
Hy-FinAgentBench (Internal) — — — — 78.5 82
IFBench 81.3 — 79.2 — — —
SWE-bench Verified — — 76.5 — 80.6 —
Humanity's Last Exam — 43.4 — — 42.7 54.9
ERQA 72.3 — — — — —
U-MATH — — — 69.8 — —
SpreadSheetBench-v1 — — — — 84.9 —
SUPERChem — — — — 62 76.7
VIBE-V2 — — — — 45 —
HorizonMath (pass@4) — — — — 4.42 5.3
Terminal-Bench 2.1 (Terminus-2) 83.3 — 80.5 — 87.9 86.7
DeepSWE — — — — 62.7 68.8
BrokenArXiv — — — — 43.1 77.7
HMMT Feb 26 — — 90.5 — — —
SocialIQA — — — 80.1 — —
Hy-FinmodelBench v2 (Internal) — — — — 51.3 66
MathArena Apex 2025 — — — — 66.3 91.4
Terminal Bench 2.1 — — 80.5 — 59.6 —
AIME 2025 — — 98.3 — — —
JobBench 55.7 — — — 54.1 68
$OneMillion-Bench — — — — 62 68.1
OSWorld 2.0 (Partial) 52.3 — — — — —
OSWorld 2.0 (Binary) 19.4 61.7 — — — 70.6
C-Eval — — — 92 — —
WideSearch — — — — 81.8 84
SimpleVQA — — — 48.5 — —
SkillsBench Avg5 — — — — 65 63.7
Vision2Web 64 — — — — 80.2
TauBench V3 Banking — — 31.27 — — 48.7
Claw-Eval Pass^3 64.4 — — — — —
E-Bench (Internal) — — — — 61.3 77.8
ProgramBench — — — — 15.5 39.5
MCP-Atlas — — 71.67 — 73.6 85.7
SWE-Marathon — — — — 19 50
Cybergym — 82.2 — — 83.3 —
Terminal-Bench 4.0 — 38.1 — — — 51.8
SWE Atlas - RF — — — — 48.6 60
DeepSWE 1.1 58.7 73.7 44.3 — — 74
FORTRESS Benign — — 95.5 — — —
Hy-CompanyBench V2 (Internal) — — — — 64.4 72.7
C³ — — — 77.3 — —
Harbor-Index — — — — 36.9 56.9
SWE Refactor Bench — 23 — — — 31
SciCode — — 39.4 — — —
AGIEval (English) — — — 75.4 — —
GDPVal-AA v2 — — — — 1580 1831
WeaveBench — — — — — 86.4
SimpleQA — — — 37.5 — —
ZEROBench — — — — — 52
Covered 27/135 16/135 21/135 29/135 56/135 76/135

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 Winner 82.8 — 76 of 135 (56.0%) STEM & Reasoning Long Context
DeepSeek V4 Pro 57.4 — 56 of 135 (42.0%) General Capabilities STEM & Reasoning
Hy4 flash 44.4 — 29 of 135 (22.0%) Knowledge Vision & Language
MiniMax-M3.1-Flash 43.7 — 16 of 135 (12.0%) Document Understanding STEM & Reasoning
Qwen3.8-Flash-Next 42.6 — 27 of 135 (20.0%) Instruction Following General Agents
Ripple-65B Preview 41.4 — 21 of 135 (16.0%) Knowledge 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 Claude Opus 5 DeepSeek V4 Pro Hy4 flash MiniMax-M3.1-Flash Qwen3.8-Flash-Next Ripple-65B Preview
General Agents 41 benchmarks 95.4 64.7 — 59.5 41.9 40.9
Agentic Coding 31 benchmarks 93.2 63.8 — 51.8 54.6 54.6
STEM & Reasoning 17 benchmarks 100 60 — 46.5 45.8 49.8
Vision & Language 9 benchmarks 85 — 41.7 67.3 66.9 —
Knowledge 8 benchmarks — — 71.4 — — 100
Long Context 3 benchmarks 59.5 — — 83.1 70 66.4
Finance 2 benchmarks 100 66.3 — — — —
General Capabilities 2 benchmarks 100 76.8 — — — —
Document Understanding 1 benchmark 95.2 — — 88.7 100 —
Instruction Following 1 benchmark — — — — 100 96.4

Partial coverage benchmarks

Scored by at least one but not all selected models (135) — missing scores show as —

Benchmark Category DeepSeek V4 Pro Qwen3.8-Flash-Next Claude Opus 5 MiniMax-M3.1-Flash Hy4 flash Ripple-65B Preview
AGIEval (Chinese) 1/6 models reasoning — — — — 90.6 —
SVG-Bench 1/6 models coding_agent 61.7 — — — — —
OSWorld-G 1/6 models general_agent — — 76.8 — — —
GPQA Diamond 4/6 models stem_reasoning 92.8 91.7 93.7 — — 85.2
CountBenchQA 1/6 models multimodal — — — — 86.6 —
OCW Courses 1/6 models stem_reasoning — — — — 64 —
ArXivMath 2/6 models stem_reasoning 62.1 — 71.5 — — —
Chinese-SimpleQA 1/6 models knowledge — — — — 74.6 —
LiveCodeBench v6 1/6 models stem_reasoning — 91.9 — — — —
Hy-LifeSearch (Internal) 2/6 models general_agent 46.9 — 56.1 — — —
Automation-Bench 3/6 models general_agent 30.4 — 48.7 62.8 — —
SWE-bench Pro 4/6 models coding_agent 55.4 62.5 79.2 — — 61.42
BFCL-V4 1/6 models general_agent — — 80.6 — — —
RealWorldQA 2/6 models vision_language — 88.5 — — 71.1 —
Agents' Last Exam (Score) 1/6 models general_agent — 51.2 — — — —
BioMysteryBench 2/6 models stem_reasoning 61.6 — 72.1 — — —
AA-LCR 2/6 models long_context — 79.1 — — — 75
MathVision 1/6 models vision_language — 90.6 — — — —
AIME 26 1/6 models stem_reasoning — — — — — 95
AndroidWorld 1/6 models general_agent — 84.5 — — — —
Agents' Last Exam 4/6 models general_agent 21.9 24.3 25.1 29.6 — —
Toolathlon Verified 3/6 models general_agent 74.1 73.5 76.5 — — —
Hy-Backend 2.0 (Internal) 2/6 models coding_agent 33.8 — 40.3 — — —
RecreationBench 1/6 models general_agent — 49.9 — — — —
Hy-SWE Max Verified (Internal) 2/6 models coding_agent 65.7 — 70.1 — — —
Hy-BrowseComp-Pro2 (Internal) 2/6 models general_agent 46.5 — 61.3 — — —
ARC-c 1/6 models reasoning — — — — 95 —
AutomationBench (Public) 1/6 models general_agent — — — — — 36.4
GDPVal 2/6 models general_agent 70.3 — 80 — — —
OfficeQA Pro 2/6 models general_agent 65.4 — 66.9 — — —
PIQA 1/6 models reasoning — — — — 84.4 —
StrongREJECT v2 (adaptive) 1/6 models safety — — — — — 97.5
SWE Atlas - TW 2/6 models coding_agent 45.6 — 68.7 — — —
MLE-bench Lite 1/6 models coding_agent — — 88 — — —
MBPP+ 1/6 models code — — — — 73.7 —
LOCA-Bench (256k) 2/6 models long_context — — 53.7 60.3 — —
WorkBuddy Blind User Evaluation (satisfaction) 1/6 models general_capabilities — — — — 2.79 —
SWE-bench Pro v2.0 Hard 1/6 models coding_agent — — — — — 79.1
SWE Atlas - QnA 2/6 models coding_agent 53.4 — 57.5 — — —
ApexBench 1/6 models multimodal_agent — — 65.8 — — —
TextVQA 1/6 models multimodal — — — — 73.4 —
NL2Repo 2/6 models coding_agent 61.5 — 75.3 — — —
MLS-Bench 1/6 models coding_agent — — 49.8 — — —
MMMU-Pro 1/6 models vision_language — — — — 52 —
Chartography 1/6 models multimodal_agent — — 84 — — —
PostTrainBench 3/6 models coding_agent 24.5 — 35 39.3 — —
INCLUDE 1/6 models multilingual — — — — 82.8 —
BabyVision 2/6 models vision_language — — 94.1 91 — —
Claw-Eval Avg 2/6 models coding_agent 58.4 60.4 — — — —
OSWorld V2.1 1/6 models general_agent — — 74 — — —
Apex-Agents 2/6 models general_agent 32.4 — 41.8 — — —
WorkSpaceBench 2/6 models general_agent 65.4 — 75 — — —
MMLU-Pro 2/6 models knowledge — — — — 71.1 84.2
Terminal-Bench Science 0.1 1/6 models coding_agent — — — 3.3 — —
BankerToolBench 3/6 models general_agent 73.1 — 81.9 76.2 — —
SuperGPQA 1/6 models knowledge — — — — 57.8 —
E-Bench-Code (Internal) 2/6 models coding_agent 64.3 — 82.7 — — —
OSWorld-Verified 1/6 models general_agent — — 86.4 — — —
CV-Bench 1/6 models vision_language — — — — 83.4 —
DeepresearchBII 1/6 models general_agent — — 54.1 — — —
Workspace-Bench-Lite 1/6 models general_agent — — 68.1 — — —
SpreadsheetBench 2 1/6 models general_agent — — 33.6 — — —
HellaSwag 1/6 models reasoning — — — — 85.2 —
HLE (with tools) 3/6 models stem_reasoning 60 35.9 60.9 — — —
LVBench 1/6 models video_understanding — 76.6 — — — —
CharXiv (RQ) 3/6 models document_understanding — 84.6 83.7 82.5 — —
WorkIndex (Tencent internal) 1/6 models general_agent — — — — 42.1 —
GDP.pdf 2/6 models long_context — — 21.6 20.8 — —
AIME 2024 1/6 models stem_reasoning — — — — — 97.1
Belebele (5-language subset) 1/6 models multilingual — — — — 93.8 —
KOR-Bench 1/6 models reasoning — — — — 54.7 —
CMMLU 1/6 models knowledge — — — — 91.6 —
CritPt (no tools) 2/6 models stem_reasoning 18 — 29.1 — — —
BrowseComp 2/6 models general_agent 83.4 — 90.8 — — —
MMLU-Redux 1/6 models knowledge — — — — 87.6 —
NL2Repo-Bench 4/6 models coding_agent 35.5 48.1 75.3 61.2 — —
DRACO 2/6 models general_agent 77.3 — 88.6 — — —
SkillsBench 1/6 models general_agent — — 63.7 — — —
WebArena-Verified 1/6 models general_agent — — 69.1 — — —
CoWorkBench 1/6 models general_agent — 73.9 — — — —
SWE-bench Multilingual 3/6 models coding_agent 77.3 81 89.5 — — —
MMLU 2/6 models knowledge — — — — 88.7 89.6
Harvey Legal Agent 1/6 models general_agent — — 22.7 — — —
PaperBench 1/6 models coding_agent — — 86.9 — — —
YC-Bench 1/6 models general_agent 500000 — — — — —
Agents' Last Exam (CLI) 1/6 models general_agent — — 29.5 — — —
Hy-FinAgentBench (Internal) 2/6 models domain_finance 78.5 — 82 — — —
IFBench 2/6 models instruction_following — 81.3 — — — 79.2
SWE-bench Verified 2/6 models coding_agent 80.6 — — — — 76.5
Humanity's Last Exam 3/6 models stem_reasoning 42.7 — 54.9 43.4 — —
ERQA 1/6 models spatial_intelligence — 72.3 — — — —
U-MATH 1/6 models stem_reasoning — — — — 69.8 —
SpreadSheetBench-v1 1/6 models general_agent 84.9 — — — — —
SUPERChem 2/6 models stem_reasoning 62 — 76.7 — — —
VIBE-V2 1/6 models coding_agent 45 — — — — —
HorizonMath (pass@4) 2/6 models stem_reasoning 4.42 — 5.3 — — —
Terminal-Bench 2.1 (Terminus-2) 4/6 models coding_agent 87.9 83.3 86.7 — — 80.5
DeepSWE 2/6 models coding_agent 62.7 — 68.8 — — —
BrokenArXiv 2/6 models stem_reasoning 43.1 — 77.7 — — —
HMMT Feb 26 1/6 models stem_reasoning — — — — — 90.5
SocialIQA 1/6 models reasoning — — — — 80.1 —
Hy-FinmodelBench v2 (Internal) 2/6 models domain_finance 51.3 — 66 — — —
MathArena Apex 2025 2/6 models stem_reasoning 66.3 — 91.4 — — —
Terminal Bench 2.1 2/6 models coding_agent 59.6 — — — — 80.5
AIME 2025 1/6 models stem_reasoning — — — — — 98.3
JobBench 3/6 models general_agent 54.1 55.7 68 — — —
$OneMillion-Bench 2/6 models general_capabilities 62 — 68.1 — — —
OSWorld 2.0 (Partial) 1/6 models general_agent — 52.3 — — — —
OSWorld 2.0 (Binary) 3/6 models general_agent — 19.4 70.6 61.7 — —
C-Eval 1/6 models knowledge — — — — 92 —
WideSearch 2/6 models general_agent 81.8 — 84 — — —
SimpleVQA 1/6 models vision_language — — — — 48.5 —
SkillsBench Avg5 2/6 models coding_agent 65 — 63.7 — — —
Vision2Web 2/6 models vision_language — 64 80.2 — — —
TauBench V3 Banking 2/6 models general_agent — — 48.7 — — 31.27
Claw-Eval Pass^3 1/6 models coding_agent — 64.4 — — — —
E-Bench (Internal) 2/6 models general_agent 61.3 — 77.8 — — —
ProgramBench 2/6 models coding_agent 15.5 — 39.5 — — —
MCP-Atlas 3/6 models general_agent 73.6 — 85.7 — — 71.67
SWE-Marathon 2/6 models coding_agent 19 — 50 — — —
Cybergym 2/6 models general_agent 83.3 — — 82.2 — —
Terminal-Bench 4.0 2/6 models coding_agent — — 51.8 38.1 — —
SWE Atlas - RF 2/6 models coding_agent 48.6 — 60 — — —
DeepSWE 1.1 4/6 models coding_agent — 58.7 74 73.7 — 44.3
FORTRESS Benign 1/6 models safety — — — — — 95.5
Hy-CompanyBench V2 (Internal) 2/6 models general_agent 64.4 — 72.7 — — —
C³ 1/6 models reasoning — — — — 77.3 —
Harbor-Index 2/6 models coding_agent 36.9 — 56.9 — — —
SWE Refactor Bench 2/6 models coding_agent — — 31 23 — —
SciCode 1/6 models — — — — — — 39.4
AGIEval (English) 1/6 models reasoning — — — — 75.4 —
GDPVal-AA v2 2/6 models general_agent 1580 — 1831 — — —
WeaveBench 1/6 models vision_language — — 86.4 — — —
SimpleQA 1/6 models knowledge — — — — 37.5 —
ZEROBench 1/6 models vision_language — — 52 — — —

Comparison FAQ

Which model scores highest in this comparison?

Claude Opus 5 scores higher overall across category strength profiles (83 vs 57 for DeepSeek V4 Pro); only 0 benchmarks are directly comparable.

Which model is strongest at Agentic Coding?

Claude Opus 5 leads the Agentic Coding category (strength score 93.2/100 across 23 comparable benchmarks).

How reliable is this comparison verdict?

Only 0 of 135 benchmarks are scored by every selected model — the verdict leans on category strength profiles and is indicative.

What now?

You've seen 6 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