LLMBenchmark.ai

Enter opens the highlighted result. Esc closes.

Claude Sonnet 4.5 vs DeepSeek V4 Flash 0423

Updated October 21, 2026. 47 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.

Replace Claude Sonnet 4.5 Replace DeepSeek V4 Flash 0423
Claude Sonnet 4.5Anthropic
Coding
59.1
Overall
20.7
Output $/M
$15.00
33
DeepSeek V4 Flash 0423DeepSeek
Coding
54.6
Overall
24.4
Output $/M
$0.182
14
vs boards won of 47 shared

Decision reading

Pick Claude Sonnet 4.5 for coding: it scores 59.1 on the Vibe Coding Index against 54.6 for DeepSeek V4 Flash 0423.

Pick DeepSeek V4 Flash 0423 when output tokens cost 82.4x less, or when the context window is larger, 1.05M against 1M tokens.

Model AClaude Sonnet 4.5Anthropic59.1/1002 independent benchmarks$1.16 per solved taskModel BDeepSeek V4 Flash 0423DeepSeek54.6/1001 independent benchmarkNo cost per solved task
Shared results
47
Claude Sonnet 4.5 only
45
DeepSeek V4 Flash 0423 only
1
Like-for-like categories
8 of 9

Shared results can support a head-to-head reading. Results only one model has describe coverage, not superiority.

Which one for your work

Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.

At a glance

Coding index
Claude Sonnet 4.5 59.1 from 2 benchmarks, DeepSeek V4 Flash 0423 54.6 from 1 benchmark.
Widest category
Coding, 13.9 points apart on average over 7 shared benchmarks.
Widest single gap
LMArena WebDev: React: Claude Sonnet 4.5 1386 against DeepSeek V4 Flash 0423 1585, 33.2 points apart on a 0 to 100 scale.
Output price
Claude Sonnet 4.5 $15.00 against DeepSeek V4 Flash 0423 $0.182 per million output tokens, 82.4x apart.
Context window
Claude Sonnet 4.5 takes 1M tokens, DeepSeek V4 Flash 0423 1.05M.

Benchmark by benchmark

First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.

Boards won, where both have a result
CategoryClaude Sonnet 4.5BarsDeepSeek V4 Flash 0423Ties
Coding43
Intelligence11
Agentic work01
Reasoning40
Math20
Knowledge31
Web design35
General163
All shared boards3314
CodingLike-for-likeFour of these boards feed the Vibe Coding Index43
Coding benchmarks for Claude Sonnet 4.5 and DeepSeek V4 Flash 0423
BenchmarkClaude Sonnet 4.5DeepSeek V4 Flash 0423Leader
SWE-bench Verified74.8%Not measuredOnly Claude Sonnet 4.5 measured
SWE-bench Multilingual67.0%Not measuredOnly Claude Sonnet 4.5 measured
LMArena WebDev13941582DeepSeek V4 Flash 0423
AA Coding Index52.152.0Claude Sonnet 4.5
LiveCodeBench (Vals AI)73.0%Not measuredOnly Claude Sonnet 4.5 measured
SWE-bench Verified (Vals AI)70.0%Not measuredOnly Claude Sonnet 4.5 measured
Vibe Code Bench (Vals AI)22.6%Not measuredOnly Claude Sonnet 4.5 measured
Terminal-Bench46.5%Not measuredOnly Claude Sonnet 4.5 measured
GSO-Bench14.7%Not measuredOnly Claude Sonnet 4.5 measured
WeirdML47.7%45.6%Claude Sonnet 4.5
Cybench60.0%Not measuredOnly Claude Sonnet 4.5 measured
ALE-BenchNot measured678.2Only DeepSeek V4 Flash 0423 measured
LMArena Text: coding14881457Claude Sonnet 4.5
LMArena Text: software and IT14731453Claude Sonnet 4.5
LMArena WebDev: React13861585DeepSeek V4 Flash 0423
LMArena WebDev: HTML13991558DeepSeek V4 Flash 0423
Neither model has a result on the other 22 boards here
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index11
Intelligence benchmarks for Claude Sonnet 4.5 and DeepSeek V4 Flash 0423
BenchmarkClaude Sonnet 4.5DeepSeek V4 Flash 0423Leader
AA Intelligence Index20.724.4DeepSeek V4 Flash 0423
Epoch Capabilities Index146.7146.1Claude Sonnet 4.5
Agentic workLike-for-likePublic boards, shown as published, kept out of the index01
Agentic work benchmarks for Claude Sonnet 4.5 and DeepSeek V4 Flash 0423
BenchmarkClaude Sonnet 4.5DeepSeek V4 Flash 0423Leader
AA Agentic Index15.826.3DeepSeek V4 Flash 0423
Design Arena: agents building web apps1060Not measuredOnly Claude Sonnet 4.5 measured
Design Arena: agents building full-stack apps1050Not measuredOnly Claude Sonnet 4.5 measured
Design Arena: agents building mobile apps1137Not measuredOnly Claude Sonnet 4.5 measured
METR time horizons67.4%Not measuredOnly Claude Sonnet 4.5 measured
Remote Labor Index2.1%Not measuredOnly Claude Sonnet 4.5 measured
GDPval42.5%Not measuredOnly Claude Sonnet 4.5 measured
DeepResearch Bench52.6%Not measuredOnly Claude Sonnet 4.5 measured
OSWorld62.9%Not measuredOnly Claude Sonnet 4.5 measured
Vending-Bench 2$3,839Not measuredOnly Claude Sonnet 4.5 measured
Neither model has a result on the other 29 boards here
ReasoningLike-for-likePublic boards, shown as published, kept out of the index40
Reasoning benchmarks for Claude Sonnet 4.5 and DeepSeek V4 Flash 0423
BenchmarkClaude Sonnet 4.5DeepSeek V4 Flash 0423Leader
GPQA Diamond81.6%Not measuredOnly Claude Sonnet 4.5 measured
MMMU Pro (Vals AI)79.3%Not measuredOnly Claude Sonnet 4.5 measured
GPQA Diamond (Epoch AI run)82.3%Not measuredOnly Claude Sonnet 4.5 measured
ARC-AGI63.7%Not measuredOnly Claude Sonnet 4.5 measured
ARC-AGI-213.6%Not measuredOnly Claude Sonnet 4.5 measured
SimpleBench54.3%46.3%Claude Sonnet 4.5
Chess puzzles10.0%Not measuredOnly Claude Sonnet 4.5 measured
Mystery game puzzles17.0%Not measuredOnly Claude Sonnet 4.5 measured
EBR-bench2.4%Not measuredOnly Claude Sonnet 4.5 measured
VPCT39.8%Not measuredOnly Claude Sonnet 4.5 measured
LMArena Text: hard prompts14621444Claude Sonnet 4.5
LMArena Text: expert14831441Claude Sonnet 4.5
LMArena Text: hard prompts in English14711449Claude Sonnet 4.5
Neither model has a result on the other 8 boards here
MathLike-for-likePublic boards, shown as published, kept out of the index20
Math benchmarks for Claude Sonnet 4.5 and DeepSeek V4 Flash 0423
BenchmarkClaude Sonnet 4.5DeepSeek V4 Flash 0423Leader
AIME88.2%Not measuredOnly Claude Sonnet 4.5 measured
MGSM (Vals AI)94.3%Not measuredOnly Claude Sonnet 4.5 measured
OTIS Mock AIME77.8%Not measuredOnly Claude Sonnet 4.5 measured
FrontierMath, tiers 1 to 323.9%Not measuredOnly Claude Sonnet 4.5 measured
FrontierMath, tier 42.4%Not measuredOnly Claude Sonnet 4.5 measured
ProofBench19.0%Not measuredOnly Claude Sonnet 4.5 measured
MATH level 597.7%Not measuredOnly Claude Sonnet 4.5 measured
LMArena Text: math14491426Claude Sonnet 4.5
LMArena Text: mathematical work14621432Claude Sonnet 4.5
Neither model has a result on the other 4 boards here
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index31
Knowledge benchmarks for Claude Sonnet 4.5 and DeepSeek V4 Flash 0423
BenchmarkClaude Sonnet 4.5DeepSeek V4 Flash 0423Leader
MMLU Pro87.4%Not measuredOnly Claude Sonnet 4.5 measured
LegalBench (Vals AI)84.1%Not measuredOnly Claude Sonnet 4.5 measured
CaseLaw (Vals AI)62.2%Not measuredOnly Claude Sonnet 4.5 measured
CorpFin (Vals AI)62.0%Not measuredOnly Claude Sonnet 4.5 measured
TaxEval (Vals AI)73.3%Not measuredOnly Claude Sonnet 4.5 measured
MortgageTax (Vals AI)64.0%Not measuredOnly Claude Sonnet 4.5 measured
MedQA (Vals AI)94.7%Not measuredOnly Claude Sonnet 4.5 measured
MedCode (Vals AI)44.1%Not measuredOnly Claude Sonnet 4.5 measured
SimpleQA Verified30.7%Not measuredOnly Claude Sonnet 4.5 measured
LMArena Text: science14481446Claude Sonnet 4.5
LMArena Text: medicine and healthcare14441448DeepSeek V4 Flash 0423
LMArena Text: legal and government14511439Claude Sonnet 4.5
LMArena Text: business and finance14401425Claude Sonnet 4.5
Neither model has a result on the other 4 boards here
Web designLike-for-likePublic boards, shown as published, kept out of the index35
Web design benchmarks for Claude Sonnet 4.5 and DeepSeek V4 Flash 0423
BenchmarkClaude Sonnet 4.5DeepSeek V4 Flash 0423Leader
Design Arena: websites11901208DeepSeek V4 Flash 0423
Design Arena: UI components11731164Claude Sonnet 4.5
Design Arena: game dev11661197DeepSeek V4 Flash 0423
Design Arena: data viz11671127Claude Sonnet 4.5
Design Arena: 3D11581189DeepSeek V4 Flash 0423
Design Arena: all code categories11831205DeepSeek V4 Flash 0423
Design Arena: SVG11141155DeepSeek V4 Flash 0423
Design Arena: ASCII art12041109Claude Sonnet 4.5
Neither model has a result on the other 1 board here
GeneralLike-for-likePublic boards, shown as published, kept out of the index163
General benchmarks for Claude Sonnet 4.5 and DeepSeek V4 Flash 0423
BenchmarkClaude Sonnet 4.5DeepSeek V4 Flash 0423Leader
SAGE (Vals AI)36.1%Not measuredOnly Claude Sonnet 4.5 measured
MedScribe (Vals AI)84.5%Not measuredOnly Claude Sonnet 4.5 measured
DTBench83.2%86.4%DeepSeek V4 Flash 0423
LMCA38.8%35.9%Claude Sonnet 4.5
LMArena Text14381432Claude Sonnet 4.5
LMArena Text: instruction following14591421Claude Sonnet 4.5
LMArena Text: creative writing14421403Claude Sonnet 4.5
LMArena Text: English14501441Claude Sonnet 4.5
LMArena Text: non-English14251419Claude Sonnet 4.5
LMArena Text: Chinese14601463DeepSeek V4 Flash 0423
LMArena Text: German14271419Claude Sonnet 4.5
LMArena Text: Japanese13901389Claude Sonnet 4.5
LMArena Text: Korean14021383Claude Sonnet 4.5
LMArena Text: French14591439Claude Sonnet 4.5
LMArena Text: Spanish14581436Claude Sonnet 4.5
LMArena Text: Russian14371427Claude Sonnet 4.5
LMArena Text: Polish14231436DeepSeek V4 Flash 0423
LMArena Text: multi-turn14651449Claude Sonnet 4.5
LMArena Text: longer queries14751434Claude Sonnet 4.5
LMArena Text: writing and language14421410Claude Sonnet 4.5
LMArena Text: entertainment and media14281400Claude Sonnet 4.5
LMArena Document1450Not measuredOnly Claude Sonnet 4.5 measured
Neither model has a result on the other 24 boards here
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
Image generation benchmarks for Claude Sonnet 4.5 and DeepSeek V4 Flash 0423
BenchmarkClaude Sonnet 4.5DeepSeek V4 Flash 0423Leader
Neither model has a result on 5 boards here

What each workload costs

Three fixed token mixes at each model's listed price, so per-token rates become a bill.

Chat turn

1K input + 500 output tokens

Claude Sonnet 4.5$0.011Fits in one request
DeepSeek V4 Flash 0423$0.0002Fits in one request

DeepSeek V4 Flash 0423 costs less

Repository review

50K input + 3K output tokens

Claude Sonnet 4.5$0.195Fits in one request
DeepSeek V4 Flash 0423$0.0051Fits in one request

DeepSeek V4 Flash 0423 costs less

Agent loop

200K input + 10K output tokens

Claude Sonnet 4.5$0.750Fits in one request
DeepSeek V4 Flash 0423$0.020Fits in one request

DeepSeek V4 Flash 0423 costs less

Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.

Specification differences

The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.

Specifications of Claude Sonnet 4.5 and DeepSeek V4 Flash 0423
SpecificationAnthropic Claude Sonnet 4.5DeepSeek V4 Flash 0423
MakerAnthropicDeepSeek
WeightsNo public weights listedOpen weights
ReleasedSeptember 29, 2025April 24, 2026
Context window1M1.05M
Input $/Mtok$3.00$0.091
Output $/Mtok$15.00$0.182
Coding index59.154.6
Independent measurements behind it21
Cost per solved task$1.16Not computable
SourceOpenRouter listingOpenRouter listing

Questions

Is Claude Sonnet 4.5 better than DeepSeek V4 Flash 0423 for coding?

Pick Claude Sonnet 4.5 for coding: it scores 59.1 on the Vibe Coding Index against 54.6 for DeepSeek V4 Flash 0423. Pick DeepSeek V4 Flash 0423 when output tokens cost 82.4x less, or when the context window is larger, 1.05M against 1M tokens.

Which is cheaper, Claude Sonnet 4.5 or DeepSeek V4 Flash 0423?

Claude Sonnet 4.5 costs $3.00 per million input tokens and $15.00 per million output tokens; DeepSeek V4 Flash 0423 costs $0.091 and $0.182. Cost per solved task is known for Claude Sonnet 4.5 only.

Which has the bigger context window, Claude Sonnet 4.5 or DeepSeek V4 Flash 0423?

DeepSeek V4 Flash 0423, with 1.05M tokens against 1M.

How many benchmarks is the Claude Sonnet 4.5 vs DeepSeek V4 Flash 0423 comparison based on?

47 benchmarks have a result for both models as of October 21, 2026. Claude Sonnet 4.5's coding index rests on 2 independent benchmarks and DeepSeek V4 Flash 0423's on 1. Where only one model has a result, the page says so instead of naming a winner.

Compare next