LLMBenchmark.ai

Enter opens the highlighted result. Esc closes.

DeepSeek V4 Pro 0813 vs GPT-5.4 Mini

Updated October 21, 2026. 66 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.

Replace DeepSeek V4 Pro 0813 Replace GPT-5.4 Mini
DeepSeek V4 Pro 0813DeepSeek
Coding
54.6
Overall
36.0
Output $/M
$1.98
63
GPT-5.4 MiniOpenAI
Coding
28.2
Overall
24.1
Output $/M
$4.50
1
vs boards won of 66 shared, 2 tied

Decision reading

Pick DeepSeek V4 Pro 0813 for coding: it scores 54.6 on the Vibe Coding Index against 28.2 for GPT-5.4 Mini.

Pick GPT-5.4 Mini when the benchmark that matters is SciCode, where it leads.

Model ADeepSeek V4 Pro 0813DeepSeek54.6/1001 independent benchmark$0.10 per solved taskModel BGPT-5.4 MiniOpenAI28.2/1001 independent benchmark$0.70 per solved task
Shared results
66
DeepSeek V4 Pro 0813 only
12
GPT-5.4 Mini only
22
Like-for-like categories
7 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
DeepSeek V4 Pro 0813 54.6 from 1 benchmark, GPT-5.4 Mini 28.2 from 1 benchmark.
Widest category
Agentic work, 16.4 points apart on average over 5 shared benchmarks.
Widest single gap
ARC-AGI-2: DeepSeek V4 Pro 0813 61.3% against GPT-5.4 Mini 18.9%, 42.4 points apart on a 0 to 100 scale.
Output price
DeepSeek V4 Pro 0813 $1.98 against GPT-5.4 Mini $4.50 per million output tokens, 2.3x apart.
Context window
DeepSeek V4 Pro 0813 takes 1.05M tokens, GPT-5.4 Mini 400K.

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
CategoryDeepSeek V4 Pro 0813BarsGPT-5.4 MiniTies
Coding1412 ties
Intelligence20
Agentic work50
Reasoning110
Math60
Knowledge80
General170
All shared boards6312 ties
CodingLike-for-likeFour of these boards feed the Vibe Coding Index141
Coding benchmarks for DeepSeek V4 Pro 0813 and GPT-5.4 Mini
BenchmarkDeepSeek V4 Pro 0813GPT-5.4 MiniLeader
LMArena WebDev15821397DeepSeek V4 Pro 0813
AA Coding Index68.856.1DeepSeek V4 Pro 0813
LiveCodeBench (Vals AI)87.5%81.5%DeepSeek V4 Pro 0813
SWE-bench Verified (Vals AI)96.4%73.0%DeepSeek V4 Pro 0813
Terminal-Bench 4 (Vals AI)14.1%2.5%DeepSeek V4 Pro 0813
Vibe Code Bench (Vals AI)82.3%48.0%DeepSeek V4 Pro 0813
Code migration (Vals AI)41.5%12.9%DeepSeek V4 Pro 0813
ProgramBench (Vals AI)0.0%0.0%Tie
IOI (Vals AI)51.6%Not measuredOnly DeepSeek V4 Pro 0813 measured
Terminal-Bench 2.1 (Vals AI)54.7%54.7%Tie
Vibe Code Bench 1-100 (Vals AI)17.5%Not measuredOnly DeepSeek V4 Pro 0813 measured
SkillsBench (Vals AI)53.8%Not measuredOnly DeepSeek V4 Pro 0813 measured
CyberBench (Vals AI)72.9%Not measuredOnly DeepSeek V4 Pro 0813 measured
SciCode51.0%52.1%GPT-5.4 Mini
Terminal-Bench Hard (Artificial Analysis run)Not measured52.3%Only GPT-5.4 Mini measured
FrontierCode28.5%27.0%DeepSeek V4 Pro 0813
WeirdML66.2%60.3%DeepSeek V4 Pro 0813
ALE-Bench1403.21188.6DeepSeek V4 Pro 0813
LMArena Text: coding14691438DeepSeek V4 Pro 0813
LMArena Text: software and IT14621436DeepSeek V4 Pro 0813
LMArena WebDev: React15881386DeepSeek V4 Pro 0813
LMArena WebDev: HTML15661419DeepSeek V4 Pro 0813
Neither model has a result on the other 16 boards here
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index20
Intelligence benchmarks for DeepSeek V4 Pro 0813 and GPT-5.4 Mini
BenchmarkDeepSeek V4 Pro 0813GPT-5.4 MiniLeader
AA Intelligence Index36.024.1DeepSeek V4 Pro 0813
Epoch Capabilities Index155.6148.7DeepSeek V4 Pro 0813
Agentic workLike-for-likePublic boards, shown as published, kept out of the index50
Agentic work benchmarks for DeepSeek V4 Pro 0813 and GPT-5.4 Mini
BenchmarkDeepSeek V4 Pro 0813GPT-5.4 MiniLeader
AA Agentic Index41.317.9DeepSeek V4 Pro 0813
SREBench (Vals AI)1.9%Not measuredOnly DeepSeek V4 Pro 0813 measured
Finance Agent (Vals AI)50.4%45.4%DeepSeek V4 Pro 0813
Tax Agent Bench (Vals AI)26.8%8.8%DeepSeek V4 Pro 0813
Legal Research Bench (Vals AI)40.9%12.5%DeepSeek V4 Pro 0813
Harvey's Legal Agent Benchmark (Vals AI)10.8%Not measuredOnly DeepSeek V4 Pro 0813 measured
Excel Modeling Benchmark (Vals AI)52.8%45.4%DeepSeek V4 Pro 0813
Terminal-Bench-Science (Vals AI)4.3%Not measuredOnly DeepSeek V4 Pro 0813 measured
τ²-Bench TelecomNot measured83.3%Only GPT-5.4 Mini measured
APEX-Agents47.3%Not measuredOnly DeepSeek V4 Pro 0813 measured
DeepResearch BenchNot measured36.3%Only GPT-5.4 Mini measured
Neither model has a result on the other 28 boards here
ReasoningLike-for-likePublic boards, shown as published, kept out of the index110
Reasoning benchmarks for DeepSeek V4 Pro 0813 and GPT-5.4 Mini
BenchmarkDeepSeek V4 Pro 0813GPT-5.4 MiniLeader
GPQA Diamond92.4%83.1%DeepSeek V4 Pro 0813
MMMU Pro (Vals AI)Not measured79.3%Only GPT-5.4 Mini measured
GPQA Diamond (Artificial Analysis run)92.8%87.5%DeepSeek V4 Pro 0813
Humanity's Last Exam (Artificial Analysis run)41.0%28.1%DeepSeek V4 Pro 0813
GPQA Diamond (Epoch AI run)91.7%86.9%DeepSeek V4 Pro 0813
ARC-AGI90.5%63.7%DeepSeek V4 Pro 0813
ARC-AGI-261.3%18.9%DeepSeek V4 Pro 0813
Chess puzzles46.0%23.0%DeepSeek V4 Pro 0813
Mystery game puzzles43.0%11.0%DeepSeek V4 Pro 0813
LMArena Text: hard prompts14581424DeepSeek V4 Pro 0813
LMArena Text: expert14591435DeepSeek V4 Pro 0813
LMArena Text: hard prompts in English14631420DeepSeek V4 Pro 0813
LMArena Vision: diagramsNot measured1266Only GPT-5.4 Mini measured
LMArena Vision: homeworkNot measured1285Only GPT-5.4 Mini measured
Neither model has a result on the other 7 boards here
MathLike-for-likePublic boards, shown as published, kept out of the index60
Math benchmarks for DeepSeek V4 Pro 0813 and GPT-5.4 Mini
BenchmarkDeepSeek V4 Pro 0813GPT-5.4 MiniLeader
AIMENot measured95.6%Only GPT-5.4 Mini measured
OTIS Mock AIME98.6%88.9%DeepSeek V4 Pro 0813
FrontierMath, tiers 1 to 364.6%51.2%DeepSeek V4 Pro 0813
FrontierMath, tier 426.8%9.8%DeepSeek V4 Pro 0813
ProofBench50.0%21.0%DeepSeek V4 Pro 0813
LMArena Text: math14411418DeepSeek V4 Pro 0813
LMArena Text: mathematical work14431429DeepSeek V4 Pro 0813
Neither model has a result on the other 6 boards here
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index80
Knowledge benchmarks for DeepSeek V4 Pro 0813 and GPT-5.4 Mini
BenchmarkDeepSeek V4 Pro 0813GPT-5.4 MiniLeader
MMLU Pro87.0%84.5%DeepSeek V4 Pro 0813
LegalBench (Vals AI)82.4%Not measuredOnly DeepSeek V4 Pro 0813 measured
CaseLaw (Vals AI)Not measured51.7%Only GPT-5.4 Mini measured
CorpFin (Vals AI)65.4%60.9%DeepSeek V4 Pro 0813
TaxEval (Vals AI)73.1%71.2%DeepSeek V4 Pro 0813
MortgageTax (Vals AI)Not measured63.5%Only GPT-5.4 Mini measured
MedCode (Vals AI)42.5%Not measuredOnly DeepSeek V4 Pro 0813 measured
SimpleQA Verified52.9%29.4%DeepSeek V4 Pro 0813
LMArena Text: science14571417DeepSeek V4 Pro 0813
LMArena Text: medicine and healthcare14461395DeepSeek V4 Pro 0813
LMArena Text: legal and government14591414DeepSeek V4 Pro 0813
LMArena Text: business and finance14421418DeepSeek V4 Pro 0813
Neither model has a result on the other 5 boards here
Web designNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
Web design benchmarks for DeepSeek V4 Pro 0813 and GPT-5.4 Mini
BenchmarkDeepSeek V4 Pro 0813GPT-5.4 MiniLeader
Neither model has a result on 9 boards here
GeneralLike-for-likePublic boards, shown as published, kept out of the index170
General benchmarks for DeepSeek V4 Pro 0813 and GPT-5.4 Mini
BenchmarkDeepSeek V4 Pro 0813GPT-5.4 MiniLeader
Vals Index53.0%Not measuredOnly DeepSeek V4 Pro 0813 measured
SAGE (Vals AI)Not measured50.8%Only GPT-5.4 Mini measured
MedScribe (Vals AI)80.2%Not measuredOnly DeepSeek V4 Pro 0813 measured
Vals Multimodal IndexNot measured54.2%Only GPT-5.4 Mini measured
IFBenchNot measured73.3%Only GPT-5.4 Mini measured
Long context reasoning (AA-LCR)80.3%77.0%DeepSeek V4 Pro 0813
DTBench93.9%80.0%DeepSeek V4 Pro 0813
LMCA45.5%40.8%DeepSeek V4 Pro 0813
LMArena Text14461412DeepSeek V4 Pro 0813
LMArena Text: instruction following14481405DeepSeek V4 Pro 0813
LMArena Text: creative writing14391370DeepSeek V4 Pro 0813
LMArena VisionNot measured1245Only GPT-5.4 Mini measured
LMArena Text: English14571412DeepSeek V4 Pro 0813
LMArena Text: non-English14311405DeepSeek V4 Pro 0813
LMArena Text: Chinese14811446DeepSeek V4 Pro 0813
LMArena Text: GermanNot measured1408Only GPT-5.4 Mini measured
LMArena Text: JapaneseNot measured1375Only GPT-5.4 Mini measured
LMArena Text: Korean14491370DeepSeek V4 Pro 0813
LMArena Text: French14461441DeepSeek V4 Pro 0813
LMArena Text: Spanish14351405DeepSeek V4 Pro 0813
LMArena Text: Russian14481417DeepSeek V4 Pro 0813
LMArena Text: PolishNot measured1415Only GPT-5.4 Mini measured
LMArena Text: multi-turn14541429DeepSeek V4 Pro 0813
LMArena Text: longer queries14581408DeepSeek V4 Pro 0813
LMArena Text: writing and language14431393DeepSeek V4 Pro 0813
LMArena Text: entertainment and media14241373DeepSeek V4 Pro 0813
LMArena Vision: EnglishNot measured1244Only GPT-5.4 Mini measured
LMArena Vision: ChineseNot measured1286Only GPT-5.4 Mini measured
LMArena Vision: reading textNot measured1253Only GPT-5.4 Mini measured
LMArena Vision: humorNot measured1235Only GPT-5.4 Mini measured
LMArena Vision: recognizing thingsNot measured1218Only GPT-5.4 Mini measured
LMArena Vision: creative writingNot measured1222Only GPT-5.4 Mini measured
Neither model has a result on the other 14 boards here
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
Image generation benchmarks for DeepSeek V4 Pro 0813 and GPT-5.4 Mini
BenchmarkDeepSeek V4 Pro 0813GPT-5.4 MiniLeader
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

DeepSeek V4 Pro 0813$0.0016Fits in one request
GPT-5.4 Mini$0.0030Fits in one request

DeepSeek V4 Pro 0813 costs less

Repository review

50K input + 3K output tokens

DeepSeek V4 Pro 0813$0.039Fits in one request
GPT-5.4 Mini$0.051Fits in one request

DeepSeek V4 Pro 0813 costs less

Agent loop

200K input + 10K output tokens

DeepSeek V4 Pro 0813$0.152Fits in one request
GPT-5.4 Mini$0.195Fits in one request

DeepSeek V4 Pro 0813 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 DeepSeek V4 Pro 0813 and GPT-5.4 Mini
SpecificationDeepSeek V4 Pro 0813OpenAI GPT-5.4 Mini
MakerDeepSeekOpenAI
WeightsOpen weightsNo public weights listed
ReleasedAugust 12, 2026March 17, 2026
Context window1.05M400K
Input $/Mtok$0.66$0.75
Output $/Mtok$1.98$4.50
Coding index54.628.2
Independent measurements behind it11
Cost per solved task$0.10$0.70
SourceOpenRouter listingOpenRouter listing

Questions

Is DeepSeek V4 Pro 0813 better than GPT-5.4 Mini for coding?

Pick DeepSeek V4 Pro 0813 for coding: it scores 54.6 on the Vibe Coding Index against 28.2 for GPT-5.4 Mini. Pick GPT-5.4 Mini when the benchmark that matters is SciCode, where it leads.

Which is cheaper, DeepSeek V4 Pro 0813 or GPT-5.4 Mini?

DeepSeek V4 Pro 0813 costs $0.66 per million input tokens and $1.98 per million output tokens; GPT-5.4 Mini costs $0.75 and $4.50. Per solved SWE-bench Verified task, DeepSeek V4 Pro 0813 costs $0.10 and GPT-5.4 Mini $0.70.

Which has the bigger context window, DeepSeek V4 Pro 0813 or GPT-5.4 Mini?

DeepSeek V4 Pro 0813, with 1.05M tokens against 400K.

How many benchmarks is the DeepSeek V4 Pro 0813 vs GPT-5.4 Mini comparison based on?

66 benchmarks have a result for both models as of October 21, 2026. DeepSeek V4 Pro 0813's coding index rests on 1 independent benchmark and GPT-5.4 Mini's on 1. Where only one model has a result, the page says so instead of naming a winner.

Compare next