LLMBenchmark.ai

Enter opens the highlighted result. Esc closes.

Claude Haiku 4.5 vs DeepSeek V3.2

Updated October 21, 2026. 62 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.

Replace Claude Haiku 4.5 Replace DeepSeek V3.2
Claude Haiku 4.5Anthropic
Coding
49.8
Overall
16.9
Output $/M
$5.00
11
DeepSeek V3.2DeepSeek
Coding
54.4
Overall
21.5
Output $/M
$0.42
50
vs boards won of 62 shared, 1 tied

Decision reading

Pick DeepSeek V3.2 for coding: it scores 54.4 on the Vibe Coding Index against 49.8 for Claude Haiku 4.5, though on 2 benchmarks against 3.

Pick Claude Haiku 4.5 when a solved task costs $0.56 against $0.99, or when the context window is larger, 200K against 164K tokens.

Model AClaude Haiku 4.5Anthropic49.8/1003 independent benchmarks$0.56 per solved taskModel BDeepSeek V3.2DeepSeek54.4/1002 independent benchmarks$0.99 per solved task
Shared results
62
Claude Haiku 4.5 only
32
DeepSeek V3.2 only
13
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 Haiku 4.5 49.8 from 3 benchmarks, DeepSeek V3.2 54.4 from 2 benchmarks.
Widest category
Coding, 6.6 points apart on average over 12 shared benchmarks.
Widest single gap
LiveCodeBench (Vals AI): Claude Haiku 4.5 41.2% against DeepSeek V3.2 80.7%, 39.5 points apart on a 0 to 100 scale.
Output price
Claude Haiku 4.5 $5.00 against DeepSeek V3.2 $0.42 per million output tokens, 11.9x apart.
Context window
Claude Haiku 4.5 takes 200K tokens, DeepSeek V3.2 164K.

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 Haiku 4.5BarsDeepSeek V3.2Ties
Coding39
Intelligence02
Agentic work01
Reasoning141 tie
Math13
Knowledge37
Web design17
General217
All shared boards11501 tie
CodingLike-for-likeFour of these boards feed the Vibe Coding Index39
Coding benchmarks for Claude Haiku 4.5 and DeepSeek V3.2
BenchmarkClaude Haiku 4.5DeepSeek V3.2Leader
SWE-bench Verified66.6%70.0%DeepSeek V3.2
SWE-bench Multilingual64.7%59.0%Claude Haiku 4.5
LiveBench coding48.1%Not measuredOnly Claude Haiku 4.5 measured
LMArena WebDev13301362DeepSeek V3.2
AA Coding Index43.944.2DeepSeek V3.2
LiveCodeBench (Vals AI)41.2%80.7%DeepSeek V3.2
SWE-bench Verified (Vals AI)66.6%67.6%DeepSeek V3.2
Vibe Code Bench (Vals AI)11.4%5.1%Claude Haiku 4.5
Code migration (Vals AI)7.5%Not measuredOnly Claude Haiku 4.5 measured
ProgramBench (Vals AI)0.0%Not measuredOnly Claude Haiku 4.5 measured
Terminal-Bench 2.1 (Vals AI)43.8%Not measuredOnly Claude Haiku 4.5 measured
LiveCodeBench (Artificial Analysis run)Not measured86.2%Only DeepSeek V3.2 measured
Terminal-Bench Hard (Artificial Analysis run)Not measured35.6%Only DeepSeek V3.2 measured
Terminal-Bench35.5%39.6%DeepSeek V3.2
WeirdML45.4%Not measuredOnly Claude Haiku 4.5 measured
LMArena Text: coding14541453Claude Haiku 4.5
LMArena Text: software and IT14391449DeepSeek V3.2
LMArena WebDev: React13261350DeepSeek V3.2
LMArena WebDev: HTML13181371DeepSeek V3.2
Neither model has a result on the other 19 boards here
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
Intelligence benchmarks for Claude Haiku 4.5 and DeepSeek V3.2
BenchmarkClaude Haiku 4.5DeepSeek V3.2Leader
AA Intelligence Index16.921.5DeepSeek V3.2
Epoch Capabilities Index142.2146.1DeepSeek V3.2
Agentic workLike-for-likePublic boards, shown as published, kept out of the index01
Agentic work benchmarks for Claude Haiku 4.5 and DeepSeek V3.2
BenchmarkClaude Haiku 4.5DeepSeek V3.2Leader
AA Agentic Index8.0Not measuredOnly Claude Haiku 4.5 measured
LiveBench research42.2%Not measuredOnly Claude Haiku 4.5 measured
LiveBench browser40.0%Not measuredOnly Claude Haiku 4.5 measured
LiveBench long agentic11.1%Not measuredOnly Claude Haiku 4.5 measured
Finance Agent (Vals AI)31.0%Not measuredOnly Claude Haiku 4.5 measured
Tax Agent Bench (Vals AI)5.9%Not measuredOnly Claude Haiku 4.5 measured
Legal Research Bench (Vals AI)10.6%Not measuredOnly Claude Haiku 4.5 measured
Excel Modeling Benchmark (Vals AI)22.7%Not measuredOnly Claude Haiku 4.5 measured
τ²-Bench TelecomNot measured90.6%Only DeepSeek V3.2 measured
APEX-AgentsNot measured21.3%Only DeepSeek V3.2 measured
BALROG31.2%Not measuredOnly Claude Haiku 4.5 measured
DeepResearch Bench45.5%Not measuredOnly Claude Haiku 4.5 measured
Vending-Bench 2$459$1,034DeepSeek V3.2
Neither model has a result on the other 26 boards here
ReasoningLike-for-likePublic boards, shown as published, kept out of the index14
Reasoning benchmarks for Claude Haiku 4.5 and DeepSeek V3.2
BenchmarkClaude Haiku 4.5DeepSeek V3.2Leader
GPQA Diamond72.2%80.3%DeepSeek V3.2
LiveBench reasoning61.7%Not measuredOnly Claude Haiku 4.5 measured
MMMU Pro (Vals AI)46.1%Not measuredOnly Claude Haiku 4.5 measured
GPQA Diamond (Artificial Analysis run)Not measured84.0%Only DeepSeek V3.2 measured
Humanity's Last Exam (Artificial Analysis run)Not measured24.6%Only DeepSeek V3.2 measured
GPQA Diamond (Epoch AI run)71.2%Not measuredOnly Claude Haiku 4.5 measured
ARC-AGI47.7%57.0%DeepSeek V3.2
ARC-AGI-24.0%4.0%Tie
Chess puzzles8.0%Not measuredOnly Claude Haiku 4.5 measured
LMArena Text: hard prompts14211434DeepSeek V3.2
LMArena Text: expert14431436Claude Haiku 4.5
LMArena Text: hard prompts in English14351446DeepSeek V3.2
Neither model has a result on the other 9 boards here
MathLike-for-likePublic boards, shown as published, kept out of the index13
Math benchmarks for Claude Haiku 4.5 and DeepSeek V3.2
BenchmarkClaude Haiku 4.5DeepSeek V3.2Leader
AIME82.7%84.6%DeepSeek V3.2
MGSM (Vals AI)92.2%86.0%Claude Haiku 4.5
AA Math IndexNot measured92.0Only DeepSeek V3.2 measured
AIME 2025 (Artificial Analysis run)Not measured92.0%Only DeepSeek V3.2 measured
OTIS Mock AIME66.7%Not measuredOnly Claude Haiku 4.5 measured
MATH level 596.4%Not measuredOnly Claude Haiku 4.5 measured
LMArena Text: math13981435DeepSeek V3.2
LMArena Text: mathematical work14301440DeepSeek V3.2
Neither model has a result on the other 5 boards here
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index37
Knowledge benchmarks for Claude Haiku 4.5 and DeepSeek V3.2
BenchmarkClaude Haiku 4.5DeepSeek V3.2Leader
MMLU Pro78.7%84.9%DeepSeek V3.2
LegalBench (Vals AI)81.2%76.1%Claude Haiku 4.5
CaseLaw (Vals AI)56.5%55.4%Claude Haiku 4.5
CorpFin (Vals AI)60.6%51.0%Claude Haiku 4.5
TaxEval (Vals AI)67.5%68.2%DeepSeek V3.2
MortgageTax (Vals AI)62.2%Not measuredOnly Claude Haiku 4.5 measured
MedQA (Vals AI)79.6%93.9%DeepSeek V3.2
MedCode (Vals AI)32.7%Not measuredOnly Claude Haiku 4.5 measured
MMLU Pro (Artificial Analysis run)Not measured86.2%Only DeepSeek V3.2 measured
SimpleQA Verified13.2%Not measuredOnly Claude Haiku 4.5 measured
LMArena Text: science14061446DeepSeek V3.2
LMArena Text: medicine and healthcare13941440DeepSeek V3.2
LMArena Text: legal and government14001425DeepSeek V3.2
LMArena Text: business and finance14001410DeepSeek V3.2
Neither model has a result on the other 3 boards here
Web designLike-for-likePublic boards, shown as published, kept out of the index17
Web design benchmarks for Claude Haiku 4.5 and DeepSeek V3.2
BenchmarkClaude Haiku 4.5DeepSeek V3.2Leader
Design Arena: websites11221175DeepSeek V3.2
Design Arena: UI components11001149DeepSeek V3.2
Design Arena: game dev10971132DeepSeek V3.2
Design Arena: data viz11241158DeepSeek V3.2
Design Arena: 3D10721133DeepSeek V3.2
Design Arena: all code categories11151164DeepSeek V3.2
Design Arena: SVG10251032DeepSeek V3.2
Design Arena: ASCII art11391082Claude Haiku 4.5
LiveBench design59.9%Not measuredOnly Claude Haiku 4.5 measured
GeneralLike-for-likePublic boards, shown as published, kept out of the index217
General benchmarks for Claude Haiku 4.5 and DeepSeek V3.2
BenchmarkClaude Haiku 4.5DeepSeek V3.2Leader
LiveBench data analysis59.3%Not measuredOnly Claude Haiku 4.5 measured
Public Benefits Bench (Vals AI)54.3%Not measuredOnly Claude Haiku 4.5 measured
SAGE (Vals AI)31.8%Not measuredOnly Claude Haiku 4.5 measured
MedScribe (Vals AI)85.2%Not measuredOnly Claude Haiku 4.5 measured
Vals Multimodal Index42.9%Not measuredOnly Claude Haiku 4.5 measured
IFBenchNot measured60.7%Only DeepSeek V3.2 measured
Long context reasoning (AA-LCR)Not measured73.3%Only DeepSeek V3.2 measured
DTBench73.6%85.6%DeepSeek V3.2
LMCA30.9%28.8%Claude Haiku 4.5
CL-benchNot measured12.4%Only DeepSeek V3.2 measured
ExploitBench13.7%Not measuredOnly Claude Haiku 4.5 measured
CL-bench LifeNot measured7.4%Only DeepSeek V3.2 measured
LMArena Text13971425DeepSeek V3.2
LMArena Text: instruction following14141413Claude Haiku 4.5
LMArena Text: creative writing13711400DeepSeek V3.2
LMArena Text: English14141439DeepSeek V3.2
LMArena Text: non-English13771409DeepSeek V3.2
LMArena Text: Chinese14181455DeepSeek V3.2
LMArena Text: German13741413DeepSeek V3.2
LMArena Text: Japanese13381374DeepSeek V3.2
LMArena Text: Korean13461370DeepSeek V3.2
LMArena Text: French14111429DeepSeek V3.2
LMArena Text: Spanish14211424DeepSeek V3.2
LMArena Text: Russian13811424DeepSeek V3.2
LMArena Text: Polish13431419DeepSeek V3.2
LMArena Text: multi-turn14091427DeepSeek V3.2
LMArena Text: longer queries14271428DeepSeek V3.2
LMArena Text: writing and language13861407DeepSeek V3.2
LMArena Text: entertainment and media13691395DeepSeek V3.2
LMArena Document1420Not measuredOnly Claude Haiku 4.5 measured
Neither model has a result on the other 16 boards here
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
Image generation benchmarks for Claude Haiku 4.5 and DeepSeek V3.2
BenchmarkClaude Haiku 4.5DeepSeek V3.2Leader
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 Haiku 4.5$0.0035Fits in one request
DeepSeek V3.2$0.0005Fits in one request

DeepSeek V3.2 costs less

Repository review

50K input + 3K output tokens

Claude Haiku 4.5$0.065Fits in one request
DeepSeek V3.2$0.015Fits in one request

DeepSeek V3.2 costs less

Agent loop

200K input + 10K output tokens

Claude Haiku 4.5$0.250Exceeds the context window
DeepSeek V3.2$0.060Exceeds the context window

DeepSeek V3.2 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 Haiku 4.5 and DeepSeek V3.2
SpecificationAnthropic Claude Haiku 4.5DeepSeek V3.2
MakerAnthropicDeepSeek
WeightsNo public weights listedOpen weights
ReleasedOctober 15, 2025December 1, 2025
Context window200K164K
Input $/Mtok$1.00$0.28
Output $/Mtok$5.00$0.42
Coding index49.854.4
Independent measurements behind it32
Cost per solved task$0.56$0.99
SourceOpenRouter listingOpenRouter listing

Questions

Is Claude Haiku 4.5 better than DeepSeek V3.2 for coding?

Pick DeepSeek V3.2 for coding: it scores 54.4 on the Vibe Coding Index against 49.8 for Claude Haiku 4.5, though on 2 benchmarks against 3. Pick Claude Haiku 4.5 when a solved task costs $0.56 against $0.99, or when the context window is larger, 200K against 164K tokens.

Which is cheaper, Claude Haiku 4.5 or DeepSeek V3.2?

Claude Haiku 4.5 costs $1.00 per million input tokens and $5.00 per million output tokens; DeepSeek V3.2 costs $0.28 and $0.42. Per solved SWE-bench Verified task, Claude Haiku 4.5 costs $0.56 and DeepSeek V3.2 $0.99.

Which has the bigger context window, Claude Haiku 4.5 or DeepSeek V3.2?

Claude Haiku 4.5, with 200K tokens against 164K.

How many benchmarks is the Claude Haiku 4.5 vs DeepSeek V3.2 comparison based on?

62 benchmarks have a result for both models as of October 21, 2026. Claude Haiku 4.5's coding index rests on 3 independent benchmarks and DeepSeek V3.2's on 2. Where only one model has a result, the page says so instead of naming a winner.

Compare next

Embed this comparison on your site

Free under CC BY 4.0, while the source link stays visible. Preview the frame · Terms and other embeds

<iframe src="https://llmbenchmark.ai/embed/compare/claude-haiku-4-5-vs-deepseek-v3-2" title="Claude Haiku 4.5 vs DeepSeek V3.2, LLMBenchmark.ai" width="100%" height="360" style="border:0;max-width:100%" loading="lazy"></iframe>
<p style="font:13px/1.4 sans-serif;margin:6px 0 0">Source: <a href="https://llmbenchmark.ai/compare/claude-haiku-4-5-vs-deepseek-v3-2">LLMBenchmark.ai, Claude Haiku 4.5 vs DeepSeek V3.2</a></p>