BenchLeader

Claude Opus 5 vs Deepseek v4 Flash Vision

Full profiles: Claude Opus 5 (Anthropic) and Deepseek v4 Flash Vision (DeepSeek): every benchmark, price and rank.

Updated · Data from 12 sources

As of 16 Nov 2026, there are not enough shared benchmarks for a verdict, and Deepseek v4 Flash Vision is 38× cheaper at $0.262 vs $10.00 per 1M tokens blended. Choose Claude Opus 5 for reasoning (66 vs 47) and multimodal (67 vs 57); choose Deepseek v4 Flash Vision for agents & tools (70 vs 64) and long context (65 vs 64), where it also streams ~1.5× faster.

Key facts (as of 16 Nov 2026)

  • Claude Opus 5 costs $10.00 per 1M tokens blended (list price, 16 Nov 2026).
  • Deepseek v4 Flash Vision costs $0.262 per 1M tokens blended (list price, 16 Nov 2026).
  • Deepseek v4 Flash Vision streams ~1.5× faster than Claude Opus 5 (Artificial Analysis, 7-day median, 16 Nov 2026).
  • Of the 16 benchmarks both models have, Claude Opus 5 leads 15 (BenchLeader, 16 Nov 2026).

Which to pick for…

The pick's numbers come first. Category scores on the BenchLeader scale (50 is the average model, 15 points one standard deviation). High confidence means a lead of half a standard deviation or more; under one point is called either.

Use casePickWhyConfidence
Agents & tool useDeepseek v4 Flash Vision70 vs 64 category score; τ³-Banking (AA) 41.0% vs 42.1%Medium
ReasoningClaude Opus 566 vs 47 category score; HLE-Diamond (AA) 44.8% vs 21.1%High
Long documentsDeepseek v4 Flash Vision65 vs 64 category score; AA-LCR v1.1 81.3% vs 79.3%Medium
Images and other mediaClaude Opus 567 vs 57 category score; MMMU-Pro 84.7% vs 74.8%High
Chat quality (human preference)–Only Claude Opus 5 has an Arena rating (68 category score)Not comparable
Lowest costDeepseek v4 Flash Vision$0.660 vs $10.00 per 1M blended; agentic coding session $0.032 vs $0.400–
Fastest first answerEither48.5 s (Claude Opus 5 (max)) vs 43.1 s (Deepseek v4 Flash Vision (max)) to the first answer token–
High volume on a budgetNeither flagshipA smaller model is usually the better fit: Cheapest good LLMs, Fastest LLMs–

Key differences

  • Not enough shared benchmarks for a verdict.
  • Claude Opus 5 (max) is stronger in multimodal, reasoning.
  • Deepseek v4 Flash Vision (max) is stronger in agents & tools, long context.
  • Deepseek v4 Flash Vision (max) is 38× cheaper ($0.262 vs $10.00 per 1M blended).
  • Deepseek v4 Flash Vision (max) streams 2.0× faster (163 vs 80 tokens per second).

At a glance

SpecClaude Opus 5Deepseek v4 Flash Vision
DeveloperAnthropicDeepSeek
Released24 Jul 202621 Aug 2026
Context window (tokens)1,000,0001,048,576
Compared atClaude Opus 5 (max)Deepseek v4 Flash Vision (max)
Open weightsNoNo
Price in / out / cached input per 1M$5.00 / $25.00 / $0.500$0.440 / $1.32 / $0.0030

Category by category

Category scores on the BenchLeader scale: 50 is the average evaluated model, 15 points is one standard deviation. A gap under one point is called level.

CategoryClaude Opus 5 (max)Deepseek v4 Flash Vision (max)Leader
Reasoning66.147.1Claude Opus 5
Coding67.0–not comparable
Agents & tools63.870.0Deepseek v4 Flash Vision
Maths69.8–not comparable
Knowledge65.2–not comparable
Human preference68.1–not comparable
Multimodal67.257.1Claude Opus 5
Long context64.365.3Deepseek v4 Flash Vision

On one chart

Each domain on the index's own scale, where 50 is an average model. Click a domain to see the benchmarks behind it. Pick up to 5 models from the list.

2550 avg75ReasoningAgents & toolsLong contextMultimodalClaude Opus 5, Reasoning: 66.1Claude Opus 5, Agents & tools: 63.8Claude Opus 5, Long context: 64.3Claude Opus 5, Multimodal: 67.2Deepseek v4 Flash Vision, Reasoning: 47.1Deepseek v4 Flash Vision, Agents & tools: 70.0Deepseek v4 Flash Vision, Long context: 65.3Deepseek v4 Flash Vision, Multimodal: 57.1

Not drawn, measured for fewer than two of these models: Coding, Maths, Knowledge, Instruction following, Human preference.

Model, click to add or removeIndex

Where sources disagree

Across the 16 benchmarks both have, Claude Opus 5 leads 15, Deepseek v4 Flash Vision leads 1.

Each publisher runs its own tasks, harnesses and effort settings, so their overall verdicts can differ; BenchLeader reconciles them on one scale. † measured at another effort setting of the same model (the publisher did not test the one compared here).

Price, speed and latency

List prices blended at three input tokens per output token. Per-task prices use the token counts on the methodology page. Cost, tokens and time per benchmark task are measured: Artificial Analysis’ averages over its whole Intelligence Index run.

MeasureClaude Opus 5 (max)Deepseek v4 Flash Vision (max)
Blended price per 1M tokens$10.00$0.660
Input / output per 1M$5.00 / $25.00$0.440 / $1.32
Cached input per 1M$0.500$0.0030
Price per task: chat reply$0.0095$0.0006
Price per task: summarize a 30-page report$0.075$0.0061
Price per task: code edit$0.068$0.0046
Price per task: agentic coding session$0.400$0.032
Price per task: structured extraction$0.015$0.0011
Agentic coding session, input mostly cached$0.198$0.012
Cost per benchmark task (AA Index)$11.82–
Output tokens per task (AA Index)118k–
Of which reasoning73k–
Time per task (AA Index)25 min–
Output speed80 tok/s163 tok/s
Time to first answer48.5 s43.1 s
Time to a complete answer, last measured 13 Nov 2026191.6 s97.8 s

Each model is compared at its best-ranked reasoning-effort setting. Claude Opus 5 (max) scores 65.0 and takes 49 s to its first answer token; Claude Opus 5 (high) scores 64.4 in 17 s, Claude Opus 5 scores 63.2 in 2 s. Deepseek v4 Flash Vision (max) and takes 43 s to its first answer token. Higher effort means more thinking before the answer: better scores, longer waits and more tokens billed. How reasoning effort changes scores and cost.

Every benchmark

144 benchmarks, grouped by category; each name links to the board and its leaders. † measured at another effort setting of the same model, shown for reference and not counted as a win.

MetricPublisherClaude Opus 5 (max)Deepseek v4 Flash Vision (max)
BenchLeader IndexBenchLeader65.0–
Agents & tools scoreBenchLeader63.870.0
Coding scoreBenchLeader67.0–
Human preference scoreBenchLeader68.1–
Knowledge scoreBenchLeader65.2–
Long context scoreBenchLeader64.365.3
Maths scoreBenchLeader69.8–
Multimodal scoreBenchLeader67.257.1
Reasoning scoreBenchLeader66.147.1
Blended price $/MBenchLeader$10.00$0.660
Output speedBenchLeader80 tok/s163 tok/s
Time to first answerBenchLeader48.5 s43.1 s
Context windowBenchLeader1,000,000 tokens1,048,576 tokens
Reasoning
GPQA DiamondEpoch AI Benchmarking Hub93.9%–
SimpleBenchSimpleBench80.6%†–
Arena Hard PromptsArena (LMArena)1513–
LiveBench 1.0 ReasoningLiveBench91.2%85.4%†
GPQA Diamond (AA)Artificial Analysis93.2%91.3%
Humanity's Last Exam (AA)Artificial Analysis54.9%34.5%
HLE-Diamond (AA)Artificial Analysis44.8%21.1%
GPQA Diamond (Vals)Vals AI93.4%†–
ARC-AGI-1ARC Prize97.5%–
ARC-AGI-2ARC Prize90.4%–
ARC-AGI-3ARC Prize30.2%†–
CritPtArtificial Analysis29.1%10.9%
CritPt v2 hard subset (AA)Artificial Analysis50.0%–
MysteryMechanismVals AI37.4%–
HLE-Diamond (Scale)Scale AI SEAL38.6%†–
Arena ExpertArena (LMArena)1530–
Chess PuzzlesEpoch AI Benchmarking Hub43.0%–
Mystery Game PuzzlesEpoch AI Benchmarking Hub59.0%–
LMCAEpoch AI Benchmarking Hub63.3%–
DTBenchEpoch AI Benchmarking Hub97.6%–
ForecastBenchForecastBench66.6%–
BTF-3Epoch AI Benchmarking Hub12.0%†–
Coding
SciCodeSciCode56.4%–
WeirdMLWeirdML91.8%–
FrontierCodeCognition53.4%–
Show all 144 benchmarks
MetricPublisherClaude Opus 5 (max)Deepseek v4 Flash Vision (max)
Coding
Arena CodingArena (LMArena)1528–
Arena WebDevArena (LMArena)1691–
LiveBench 1.0 CodingLiveBench81.5%68.2%†
SciCode (AA)Artificial Analysis56.4%49.6%
LiveCodeBenchVals AI89.0%–
IOIVals AI84.3%–
SWE-bench (Vals)Vals AI97.0%†–
Code MigrationVals AI57.5%–
ProgramBenchVals AI3.0%–
Vibe Code Bench 1-100Vals AI28.5%–
Vibe Code Bench v1.1Vals AI88.4%†–
SWE Atlas v1.0: Codebase QnAScale AI SEAL63.2%†–
SWE Atlas v1.0: Test WritingScale AI SEAL62.2%†–
SWE-Bench Pro V2Scale AI SEAL99.4%†–
SWE-Bench Pro V2 HARDScale AI SEAL98.0%†–
MirrorCodeEpoch AI Benchmarking Hub87.8%–
DeepSWE v1.1Epoch AI Benchmarking Hub73.7%–
CursorBenchCursor46.6%–
ALE-BenchEpoch AI Benchmarking Hub2164.6†–
FrontierSWEEpoch AI Benchmarking Hub52.0%–
WeirdML v3WeirdML19.0%†–
Terminal-Bench 4.0 (AA)Artificial Analysis49.0%12.1%
Terminal-Bench 2.1 (AA)Artificial Analysis89.1%74.2%
Startrise FrontendStartrise82.3%†–
GameBench 2 OverallDuelLab GameBench 276.5†–
GameBench 2 by reasoning settingDuelLab GameBench 284.8†–
Design Arena WebsiteDesign Arena1304†–
FrontierCode 2.0 (AA)Artificial Analysis31.1%–
Agents & tools
Terminal-BenchTerminal-Bench52.1%–
OSWorld-Verified 2.0OSWorld31.4%–
APEX-AgentsMercor65.8%–
Arena AgentArena (LMArena)7.2–
LiveBench Agentic CodingLiveBench65.2%65.1%†
Terminal-Bench 2.1 (Vals)Vals AI84.6%†–
MCP AtlasScale AI SEAL85.8%†–
HiL-BenchScale AI SEAL57.0%†–
τ³-bench Bankingτ²-bench48.7%–
GDPval-AA v2.1, normalised scoreArtificial Analysis–52.3%
GDPval-AA v2.1Artificial Analysis17291555
τ³-Banking (AA)Artificial Analysis42.1%41.0%
Analyst Agent (AA)Artificial Analysis53.8%–
CUA-benchVals AI9.0%–
CyberBenchVals AI65.4%–
Finance Agent v2Vals AI58.6%–
Harvey's Legal Agent Benchmark v1Vals AI6.7%–
Legal Research BenchVals AI55.3%–
SkillsBenchVals AI60.4%–
SREBenchVals AI12.2%–
Tax Agent BenchVals AI47.1%–
Terminal-Bench 4.0 (Vals)Vals AI53.5%–
Terminal-Bench ScienceVals AI27.1%–
Time Horizon Index: KSPVals AI18.8%–
Arena SearchArena (LMArena)1213†–
Arena Agent Task CompletionArena (LMArena)10.2–
BALROGBALROG63.4%–
PostTrainBenchEpoch AI Benchmarking Hub35.7%†–
GBAEvalEpoch AI Benchmarking Hub79.6%†–
Vending-Bench 2Andon Labs11181.9†–
AutomationBenchArtificial Analysis56.6%47.5%
GDP.pdf (AA)Artificial Analysis21.6%12.0%
EnterpriseOps-GymArtificial Analysis47.5%47.1%
AA-Briefcase v1.1Artificial Analysis16541418
Design Arena AgenticDesign Arena1270†–
Terminal-Bench-Science 0.1 (AA)Artificial Analysis28.6%–
DeepSec-Bench (AA)Artificial Analysis25.9%–
Maths
FrontierMath Tiers 1–3Epoch AI Benchmarking Hub85.6%–
FrontierMath Tier 4Epoch AI Benchmarking Hub73.2%–
OTIS Mock AIMEEpoch AI Benchmarking Hub98.9%–
ProofBenchVals AI99.0%–
LiveBench MathematicsLiveBench95.7%87.8%†
Arena MathsArena (LMArena)1515–
FrontierMath ErdősEpoch AI Benchmarking Hub0.0%–
Knowledge
SimpleQA VerifiedEpoch AI Benchmarking Hub59.9%–
LiveBench 1.0 Data AnalysisLiveBench74.5%79.5%†
AA-Omniscience v1.1Artificial Analysis37.6–
MMLU-ProVals AI91.6%–
LegalBenchVals AI87.0%–
CorpFinVals AI73.2%†–
TaxEvalVals AI75.1%†–
BioMysteryBenchVals AI79.3%–
Excel Modeling BenchmarkVals AI73.6%–
MedCodeVals AI63.6%–
MedScribeVals AI91.0%–
Public Benefits Bench v1.1Vals AI76.9%–
DrugDiscoveryBenchScale AI SEAL53.7%–
EBR-benchEpoch AI Benchmarking Hub45.7%–
GDP.pdfEpoch AI Benchmarking Hub24.0%–
AA-Omniscience v1.1: accuracyArtificial Analysis60.9%–
AA-Omniscience v1.1: non-hallucinationArtificial Analysis40.6%–
FACTS ParametricGoogle48.8%†–
Instruction following
LiveBench LanguageLiveBench88.7%80.4%†
LiveBench Instruction FollowingLiveBench63.8%71.0%†
Arena Instruction FollowingArena (LMArena)1494–
Human preference
Arena TextArena (LMArena)1489–
EQ-Bench 4EQ-Bench1365†–
EQ-Bench Creative Writing v3EQ-Bench2135†–
Arena Creative WritingArena (LMArena)1473–
Arena Multi-turnArena (LMArena)1486–
Arena Longer QueriesArena (LMArena)1502–
Multimodal
Arena VisionArena (LMArena)1318†–
MMMU-ProArtificial Analysis84.7%74.8%
MMMU-Pro (Vals)Vals AI89.9%–
MortgageTaxVals AI72.1%†–
Humanity's Sixth SenseScale AI SEAL30.5%–
Blueprint-Bench 2Epoch AI Benchmarking Hub30.4%†–
Furniture AssemblyEpoch AI Benchmarking Hub60.8%–
Long context
AA-LCR v1.1Artificial Analysis79.3%81.3%
Arena DocumentArena (LMArena)1486†–
MLCRArtificial Analysis11.1%–
Composite
Epoch Capabilities IndexEpoch AI Benchmarking Hub164.7†–
LiveBench 1.0LiveBench80.1%76.8%†
AA Intelligence Index v5.0.1Artificial Analysis44.2–
AA Intelligence Index v4.3.2Artificial Analysis50.834.8
AA Progress Index v1Artificial Analysis333.7308.5
Vals RSI IndexVals AI33.0%–
Vals Science IndexVals AI47.9%–
Vals Multimodal IndexVals AI73.9%†–
Safety & honesty
FORTRESSScale AI SEAL16.8%†–
DistressBenchScale AI SEAL70.2%†–

Data as of 2026-11-16. Best configuration of each model; every score links to its source on the model pages.

DuelLab GameBench 2, release gb2-20261111-7b598bb13de4 (https://benchmarks.duellab.org/releases/gb2-20261111-7b598bb13de4/), CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Scores mapped to BenchLeader model names; not endorsed by DuelLab.

What changed

  • Claude Opus 5 (max): 25 new results (LiveCodeBench, MMLU-Pro, IOI…).
  • Claude Opus 5 (max): 1 new result (GDPval-AA v2.1).
  • Claude Opus 5 (max): 1 new result (MirrorCode).
  • Claude Opus 5 (max): BenchLeader Index 65.5 → 65.0.
  • Claude Opus 5 (max): 7 new results (AA Intelligence Index v5.0.1, HLE-Diamond (AA), AA-Omniscience v1.1…).
  • Claude Opus 5 (max): 1 new result (MirrorCode).
  • Claude Opus 5 (max): 1 score revision.
  • Claude Opus 5 (max): BenchLeader Index 66.3 → 65.5.
  • Claude Opus 5 (max): 1 new result (AA Progress Index v1).
  • Deepseek v4 Flash Vision (max): 2 new results (AA Progress Index v1, HLE-Diamond (AA)).

Rankings: Best LLM overall · Best LLM for agents · Best cheap LLMs

Claude Opus 5 vs Deepseek v4 Flash Vision: questions

Claude Opus 5 vs Deepseek v4 Flash Vision: which is cheaper, and will that change?
Deepseek v4 Flash Vision is cheaper today: $0.262 against $10.00 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
Which is faster, Claude Opus 5 or Deepseek v4 Flash Vision?
Deepseek v4 Flash Vision streams ~1.5× faster (163 vs 80 output tokens per second, Artificial Analysis 7-day median).
Is Claude Opus 5 or Deepseek v4 Flash Vision better for agents and tool use?
Deepseek v4 Flash Vision scores higher in agentic tasks (70 vs 64 on the category index, where 50 is average). On τ³-Banking (AA), the top shared agents & tools benchmark, it is 42.1% for Claude Opus 5 against 41.0% for Deepseek v4 Flash Vision.
Which is better for long documents?
Deepseek v4 Flash Vision scores higher on long-context benchmarks (65 vs 64). Both take about 1M tokens: Claude Opus 5 1,000,000, Deepseek v4 Flash Vision 1,048,576.
Which has lower latency?
Deepseek v4 Flash Vision answers first: 43.1 s against 48.5 s to the first answer token, as Artificial Analysis measures it. Reasoning effort drives most of this: a higher setting thinks longer before it answers.
More questions (2)
Do independent benchmarks agree on Claude Opus 5 vs Deepseek v4 Flash Vision?
Across the 16 benchmarks both have, Claude Opus 5 leads 15, Deepseek v4 Flash Vision leads 1.
When was Deepseek v4 Flash Vision released?
Deepseek v4 Flash Vision was released on 21 Aug 2026.